AI Coding Tools Pricing Comparison 2026

As of September 28, 2026, the paid entry point for every major AI coding agent is still about $20 a month: Claude Pro ($20, with Claude Code), ChatGPT Plus ($20, with Codex), and Cursor Pro ($20), while GitHub Copilot Pro stays at $10 and Google Antigravity has a free individual tier. The heavy-use tiers sit at $100 to $200: Claude Max ($100 or $200), ChatGPT Pro (from $100, with new $200 sign-ups currently paused), Cursor Ultra ($200), and Copilot Max ($100). The biggest real differences are in how usage is metered, not the sticker price.

Last updated: September 28, 2026. Claude, Codex, Cursor, GitHub Copilot, Antigravity, and Claude Cowork prices were re-verified against their official pages that day - see What Changed on September 28. Windsurf, Devin, Augment, Kiro, Zed, and the app builders were not re-checked in this pass and carry their earlier verification dates. Verify any plan detail against the official pricing page before you buy.

Official Sources#

The AI coding tool market has more options and more pricing complexity than ever. Some tools are free. Some cost $200 a month. Some charge per task or per credit. Figuring out what you actually get for your money means digging through pricing pages, fair-use policies, and fine print that changes monthly. If you are choosing by workflow first, the AI coding tools comparison matrix is the companion piece to keep open.

Previous update (August 27, 2026): Claude Sonnet 5's $2/$10 per MTok rate is now the permanent standard price: Anthropic confirmed on its pricing page (August 15, 2026) that the previously scheduled September 1 increase to $3/$15 will not occur, so any "promo ends August 31" framing in the July sections below is outdated. Anthropic Opus 5 is confirmed with Fast Mode at 2x standard pricing. Opus 4.8 now listed as a legacy model on the official pricing page. OpenAI's GPT-5.6 Sol dropped to $4/$20 per MTok on the API (from $5/$30), with promotional pricing through at least November 21, 2026. All prices verified August 27, 2026, except Windsurf and Devin rows, which were rate-limited during this check and still carry their July 27 verification. Verify any plan detail against the official pricing page before you buy.

For the newest numbers, see the Recent Pricing Changes section below, and for how the Fable 5 tier shifts cost-per-task math, the Fable 5 cost analysis. If you run agent fleets rather than a single session, the economics of Fable 5 orchestrators and Sonnet 5 workers works through the one-expensive-manager, many-cheap-workers math.

If you only need the fastest decision path:

• Compare workflow fit first: AI coding tools comparison matrix
• Compare the top 3 agent styles: Claude Code vs Cursor vs Codex
• Want a ranked shortlist first: /best/ai-coding-tools
• Jump straight to commercial intent: /pricing and /compare

This is the complete breakdown. Every major AI coding tool, what each tier costs, what it includes, and where the hidden costs live. Core source links re-checked September 28, 2026. If you want to plug your own usage numbers in before you commit, run them through our AI cost calculator and you will get a per-month estimate across providers in a few seconds.

If you are comparing tools rather than just reading the sticker price, keep three companion pages open: the AI coding tools comparison matrix, and the Claude Code vs Cursor vs Codex breakdown. Those pages go deeper on tradeoffs, recent plan changes, and which tool fits which workflow.

Pricing Sources and Freshness#

Pricing changes quickly, especially for tools that mix subscriptions, credits, premium model pools, and usage-based billing. Treat these as checked public prices, not permanent guarantees. For purchase decisions, verify the official pricing page before you buy.

For tool selection after budget, use best AI coding tools 2026.

What Changed on September 28#

A full re-verification of the tools with the most pricing search interest. Everything in this section was checked against official pages on September 28, 2026.

Claude Code (Anthropic). Claude Code is included in Pro and Max, not Free.

On the API side, Anthropic's pricing page now lists Opus 5.5 at $4/$20 per MTok (with Fast Mode at 2x standard pricing), Fable 5.1 at $10/$50, Sonnet 5 at $2/$10, and Haiku 4.5 at $1/$5. Opus 5 has moved to the legacy list at $5/$25. On consumer plans, Fable is paid with usage credits on Pro and capped at 50% of weekly limits on Max. See the Opus 5.5 release guide for what the new default model changes.

Claude Cowork. There is no separate Cowork price. The Cowork page says it works with a paid plan, and Anthropic's banner reads "Claude Cowork is now just Claude," rolling out to Pro and Max first. Team seats list "Claude Code and Claude Cowork" together. Cowork runs on your plan's usage, so heavy Cowork use can push you toward Max. We break down which one to reach for in Claude Cowork vs Claude Code.

OpenAI Codex. Codex is bundled into ChatGPT plans, and ChatGPT Work now shares the same usage limits:

Two things to know before you buy: OpenAI's help center says new sign-ups and upgrades to the Pro $200 plan are temporarily paused (existing subscriptions keep renewing), and GPT-5.5 retires from ChatGPT and Codex on October 14, 2026. OpenAI publishes estimated local messages per five-hour window by model; for example, GPT-6 Sol is listed at 15 to 150 on Plus versus 300 to 3,000 on Pro 20x. Our Codex usage limits and pricing guide walks through those windows in detail.

Cursor (and Composer). Plan prices are unchanged: Hobby free, Pro $20, Pro Plus $60, Ultra $200, Teams $40 Standard or $120 Premium. What changed is the Cursor Models pool, which now covers Grok 4.7, Grok 4.6, Grok 4.5, and Composer 2.5. Per Cursor's model pricing docs, Composer 2.5 is $0.50 input and $2.50 output per MTok ($3/$15 in Fast mode), far below third-party frontier models in the Other Models pool, such as Claude Opus 5.5 at $4/$20. Cursor also lists a Start plan for developers in India at 649 rupees a month that covers only the Cursor Models pool. Cursor's own guidance: daily agent users typically spend $60 to $100 a month in total usage. See the Composer 2.5 developer guide for how the model performs.

GitHub Copilot. Free ($0, 2,000 completions a month), Pro $10 ($15 in monthly AI credits), Pro+ $39 ($70), Max $100 ($200). Business is $19 per seat and Enterprise $39 per seat. New on the plan page: Pro includes access to third-party agents, naming Claude Code and Codex.

Google Antigravity. Antigravity has a free individual tier with unlimited tab completions and weekly rate limits, using Gemini 3.8 Flash, Gemini 3.1 Pro, Claude Sonnet and Opus 4.6, and gpt-oss-120b. Higher limits come from Google AI plans: Google AI Pro at $19.99/mo and Google AI Ultra from $99.99/mo (well below the $250 Ultra price this post listed earlier). Pro and Ultra can buy AI credits for overage. Organizations get Antigravity through Google Cloud with consumption-based pricing. For how it compares as an agent, see Antigravity CLI vs Claude Code vs Codex.

September 2026 price table for the big six#

What Changed on August 27#

• Claude Sonnet 5's promotional pricing is now permanent. Anthropic's pricing page (verified August 27, 2026) lists Sonnet 5 at $2 input / $10 output per MTok with no end date, and explicitly states the previously scheduled September 1 increase to $3/$15 will not occur. Every "Sonnet 5 promo ends August 31" note in the July sections of this post is outdated - treat $2/$10 as the going rate. Opus 5 remains at $5/$25 with Fast Mode at 2x ($10/$50), and Opus 4.8 stays listed as a legacy model.
• OpenAI cut GPT-5.6 Sol to $4/$20 per MTok. The API pricing page now lists Sol at $4 input / $20 output (cached $0.40), down from $5/$30, with long-context rates of $8/$30 and promotional pricing through at least November 21, 2026. The lower rate also flows through to Codex and ChatGPT Work usage counting, stretching subscription quotas further.
• Cursor restructured individual limits into two usage pools. Per Cursor's pricing docs, Pro/Pro+/Ultra each include a Cursor Models pool (Grok 4.6, Grok 4.5, Composer 2.5) plus an Other Models pool charged at the selected model's API rate, with on-demand usage billed at the same rates. Prices are unchanged: Pro $20, Pro+ $60, Ultra $200, Teams Standard $40, Premium seat $120.
• GitHub Copilot Pro+ is now a permanent part of the individual lineup at $39/mo with 4x+ Pro's included usage and $70 in monthly credits, between Pro ($10) and Max ($100).

All prices verified August 27, 2026. Windsurf and Devin pricing pages were rate-limited during this check, so their rows still carry July 27 verification - verify those before purchasing.

What Changed on July 27#

• Kimi K3 open weights now available. Moonshot AI delivered the full 2.8T-parameter model weights and inference code on Hugging Face and ModelScope as promised July 27. The model already has 5.7k likes and 2,850 downloads on Hugging Face. API pricing remains $3/$15 per MTok. See the Kimi K3 developer guide for architecture details and the access comparison post for every provider route.
• Anthropic Opus 5 Fast Mode confirmed at 2x pricing. The official pricing page now lists Opus 5 with Fast Mode at 2x standard pricing ($10/$50 per MTok) for workloads that need up to 2.5x faster speeds, alongside the standard $5/$25 per MTok rate. Prompt caching is confirmed at $6.25/$0.50 write/read.
• Opus 4.8 now marked as legacy model. Anthropic's pricing page now lists Opus 4.8 under the "Legacy models" section. Pricing unchanged ($5/$25 per MTok) but the labeling shift signals deprecation direction. Opus 5 is the recommended replacement at the same price point with better reasoning.
• Claude Sonnet 5 promotional pricing confirmed. The $2/$10 per MTok intro rate continues through August 31, 2026. Standard pricing after September 1 will be $3/$15 per MTok. Confirmed on the official pricing page July 27.

All prices verified July 27, 2026.

What Changed on July 24#

• Kimi K3 open weights delayed to July 27. Moonshot AI announced a three-day delay from the original July 24 release target. The full 2.8T-parameter model weights and inference code would be available on Hugging Face and ModelScope on July 27. API pricing remains $3/$15 per MTok. Early access partners reported 15-20% speed improvements over the initial launch build.
• Claude Sonnet 5 promotional pricing status confirmed. Anthropic reconfirmed the $2/$10 per MTok intro rate continues through August 31, 2026. After September 1, expect $3/$15 per MTok. Current Sonnet 5 usage counts against plan limits during the promo period, not usage credits.
• GPT-5.6 capacity updates. OpenAI reported infrastructure improvements that stabilized GPT-5.6 Sol tier access for all Plus and Pro users. The temporary five-hour usage limit from July 13-15 is now permanently removed. Sol tier response times improved 12% on average across coding workloads.

What Changed on July 17#

• Kimi K3 launched with 2.8T parameters. Moonshot AI released Kimi K3, a 2.8-trillion-parameter Mixture-of-Experts model with native vision and a 1M-token context window. API pricing is $3 per million input tokens, $0.30 cached, and $15 per million output tokens. Full weights promised by July 27. See the Kimi K3 developer guide for architecture details and cost math.
• Gemini 3.5 Pro missed its July 17 target. Google's rebuilt Gemini 3.5 Pro missed its third consecutive deadline. The model remains in limited enterprise preview on Vertex AI. Polymarket now shows 81% probability of a July 31 delivery. The public Gemini API does not list a 3.5 Pro model ID.
• Fable 5 countdown: two days left. Pro, Max, Team, and Enterprise subscribers have until Saturday July 19 at 11:59 PM PT to use Fable 5 at included rates. Starting Sunday July 20, all Fable 5 usage runs on usage credits at $10/$50 per MTok.

All prices verified July 17, 2026.

What Changed on July 15#

• Codex and ChatGPT Work hit 8 million active users. OpenAI reported the milestone on July 14-15, up from 1 million in February and 6 million just three days earlier. Sam Altman called GPT-5.6 Sol growth "insane" and cited a 2.5x spike in agentic product usage over the prior week.
• OpenAI keeping GPT-5.6 Sol limits lifted. OpenAI said it will continue operating without the five-hour rate limit through the user surge, giving all plans more access to GPT-5.6 Sol across Codex and ChatGPT Work.
• Fable 5 deadline unchanged at July 19. Four days remain at included rates. Pro, Max, Team, and Enterprise subscribers can use Fable 5 for up to 50% of weekly limits until 11:59 PM PT on Saturday. Starting July 20, all Fable 5 usage runs on usage credits at $10/$50 per MTok.

All prices verified July 17, 2026.

What Changed on July 14#

• Fable 5 plan access extended to July 19. Anthropic announced the third extension on July 12, giving Pro, Max, Team, and Enterprise subscribers another week of Fable 5 at included rates. The deadline is now Sunday, July 19 at 11:59 PM PT. After that, Fable 5 transitions to usage-based billing at $10/$50 per MTok. This represents the third delay since the original June 22 cutoff as Anthropic works to secure capacity.
• OpenAI lifted GPT-5.6 Sol usage limits temporarily. Following intense demand after the Codex and ChatGPT Work launches, OpenAI removed the five-hour usage restriction for Plus, Pro, and Business plans on July 13. Usage counters were reset for all users. OpenAI also rolled out efficiency improvements to GPT-5.6 Sol that reduce token consumption while handling the same workload.
• Kimi K2.7 Code now on Copilot Business and Enterprise. Moonshot AI's K2.7 Code model, previously available only to Copilot Pro, Pro+, and Max plans, is now additionally available on Copilot Business and Copilot Enterprise tiers.

All prices verified July 17, 2026.

What Changed on July 12#

• GPT-5.6 global rollout continues. OpenAI is rolling out GPT-5.6 across ChatGPT, Codex, and the API globally. The three-tier Sol family is now more widely available: Sol ($5/$30 per MTok) for frontier reasoning, Terra ($2.50/$15) as a balanced everyday model, and Luna ($1/$6) as the fastest option. CEO Sam Altman noted Sol is 54% more token efficient on agentic coding tasks.
• GitHub Copilot adds session spend limits. You can now set AI credit session limits in Copilot CLI and the GitHub Copilot SDK to cap the amount an agent spends in a single session. This addresses the unpredictable credit consumption that hit some teams after the June billing transition.
• Copilot agentic browser tools GA. Browser automation tools for Copilot agents are now generally available by default, letting agents navigate pages, inspect content, capture screenshots, and validate web apps directly in VS Code.

All prices verified July 12, 2026.

What Changed on July 11#

• ChatGPT Work launched. OpenAI released ChatGPT Work, powered by GPT-5.6, to turn scattered notes and drafts into finished work. It brings together context from your team's tools with multi-agent capabilities that run concurrent subagents. Available to all ChatGPT plans on desktop, rolling out to web and mobile over the next few days. GPT-5.6 in Work uses Max mode by default; Ultra mode is available to Pro and Enterprise users.
• GPT-5.6 Sol family pricing confirmed. The three-tier Sol family is now in limited preview with confirmed API pricing: Sol ($5/$30 per MTok) for frontier reasoning, Terra ($2.50/$15) as a balanced everyday model at 2x lower cost than GPT-5.5, and Luna ($1/$6) as the fastest and most affordable option.
• Claude Code 2.1.207 released. Auto mode is now available without opt-in on Bedrock, Vertex AI, and Foundry. Bedrock, Vertex, and Claude Platform on AWS default to Claude Opus 4.8. Key fixes include terminal freezing during long streams, spurious prompt-injection warnings, and Bedrock credential handling. Shell-injection hardening blocks ${user_config.*} in plugin hook commands.
• Claude Monthly Recap feature. A new Reflect feature in Settings shows topics you spent time on, your most active day and peak hour. Available in beta across Free, Pro, and Max plans with memory enabled.

What Changed on July 9#

• Fable 5 is now API-only. The promotional window that restored Fable 5 to Pro/Max/Team plans ended July 7. From July 8 onward, Fable 5 access requires API billing at $10/$50 per MTok. Route premium coding work through Sonnet 5 ($2/$10 intro pricing) or Opus 4.8 ($15/$75) if you want to stay on your subscription.
• Claude Cowork expanded to web and mobile. Claude Cowork - the agentic work assistant with Microsoft 365 write tools - is now available on web and mobile in addition to desktop, rolling out to Max plan users first. Claude can now draft, send, and organize email, manage calendar events, and create files in OneDrive and SharePoint.
• Claude Code 2.1.205 released July 8. Key reliability fixes: auto mode now blocks tampering with session transcript files, fixed --json-schema silently failing on invalid schemas, fixed messages sent while Claude was working being lost at --max-turns limit. Full changelog at code.claude.com/docs/en/changelog.
• Claude for Government public beta. Claude Code and Claude Cowork are now in public beta for Claude for Government Desktop - FedRAMP High authorized AI with tamper-evident audit logs and spend governance.
• Meta Muse Spark 1.1 API launched. Meta opened its first paid AI model to developers through the new Meta Model API. Muse Spark 1.1 offers a 1M token context window, tool calling, and MCP support at $1.25/$4.25 per MTok - significantly cheaper than Claude or GPT for high-volume agentic workloads. US developers get $20 in free credits. See our Meta Muse Spark 1.1 developer guide for the full breakdown.

All prices verified July 9, 2026.

What Changed on July 8#

• Cursor Teams pricing overhaul. Cursor split Teams seat usage into two separate pools: one for Composer and Auto (first-party Cursor models), and one for third-party API models (Claude, GPT, Gemini). Standard seats stay at $40/mo with more total usage. A new Premium seat ($120/mo) offers 5x Standard usage for power users running heavy agent workloads. Cursor estimates the changes lower costs for 90% of teams.
• VS Code 1.128 released. Multi-chat support for Claude agent sessions lets you run related conversations in parallel, compare approaches, and branch from earlier turns. Copilot Vision is now generally available in VS Code Chat - attach images and PDFs by pasting, dragging, or dropping them. See our VS Code 1.128 developer guide.
• GPT-5.6 Sol family still in limited preview. Sol ($5/$30), Terra ($2.50/$15), and Luna ($1/$6) remain available only to selected API and Codex partners. No general availability date announced yet.
• Sonnet 5 introductory pricing continues. The $2/$10 per MTok intro rate runs through August 31, 2026. After that, expect $3/$15 per MTok.

What Changed on July 7#

• Fable 5 promotional window ended. The deadline hit on July 7 - Fable 5 is now API-only at $10/$50 per MTok.
• OpenAI previewed GPT-5.6 Sol. The three-tier Sol family is in limited preview: Sol for frontier reasoning and long-horizon agentic work, Terra as a balanced everyday model at 2x lower cost than GPT-5.5, and Luna as the fastest and most affordable option. See our GPT-5.6 Sol developer guide for the full breakdown.
• ChatGPT Pro pricing restructured. ChatGPT Pro now offers $100/mo with 5x usage or $200/mo with 20x usage. The Go tier moved from $6 to $8/mo. Business is $20/user/mo on annual billing ($25 monthly). GPT-5.5 Instant is the default model across all tiers.
• Claude Code default permissions changed. The default permission mode is now Manual across CLI, VS Code, and JetBrains. This affects new installs and fresh sessions.

What Changed on July 6#

• Fable 5 final countdown. After the June 30 US export control lift, Fable 5 was restored to Pro/Max/Team plans at 50% limits through July 7.
• Sonnet 5 became default. As of July 4, Claude Sonnet 5 replaced Sonnet 4.6 as the default model.
• Copilot Vision is GA. GitHub Copilot's vision capabilities moved to general availability as of July 4. Premium model selections draw from your AI Credit pool.
• Kimi K2.7 Code added to Copilot. Moonshot AI's K2.7 Code model is now available in GitHub Copilot's model roster.
• Cursor iOS beta launched. 75% off iOS promo ended July 5. Cursor Teams now includes MCP support.

Recent Pricing Changes (June 2026)#

The June 2026 pricing wave moved several tools at once:

• GitHub Copilot usage-based billing is live. As of June 1, 2026, all Copilot plans transitioned to usage-based billing with AI Credits. Pro includes $15/month in credits, Pro+ includes $70/month, and Max includes $200/month. Premium model selections draw from your credit pool. See GitHub Copilot's new usage-based billing for the full breakdown.
• Cursor credit system confirmed. Cursor's credit-based model remains in effect. Auto mode is unlimited; manually selecting premium models like Claude Sonnet or GPT-4 draws from your monthly credit pool equal to your plan price.
• Sonnet 4.6 launched June 17. API pricing reflects Sonnet 4.6 at $3/$15 per MTok, which replaced Sonnet 4.5 as the default Sonnet-tier model. Sonnet 4.5 remains available at the same price but is now listed under legacy models in Anthropic's docs. (Now superseded by Sonnet 5 as of July 4.)

Where Pricing Fits in the Decision Web#

Pricing should be the second filter, not the first. Use the adjacent guides this way:

• Start with what an AI coding agent is if you are still separating autocomplete, IDE agents, terminal agents, and cloud agents.
• Use the comparison matrix when you need feature-by-feature tradeoffs.
• Use Claude Code vs Cursor vs Codex for the three-tool decision most teams actually make.
• Use Cursor vs Codex when the choice is IDE-native iteration versus OpenAI's agent stack.
• Use Anthropic vs OpenAI developer experience when the decision is really platform strategy, not editor choice.

That web prevents the common mistake: picking a cheaper plan for a workflow the tool was never built to handle.

Master Pricing Table#

Now the details on every tool.

Claude Code#

Claude Code is Anthropic's terminal-native AI agent. No IDE, no editor. It reads your entire codebase, edits files, runs commands, and operates autonomously across your whole project. The reasoning quality on complex tasks is the best in class. Start with what Claude Code is, then compare it against Cursor, Codex, and Aider if you are choosing a primary agent.

Pro ($20/mo, or $17/mo billed annually): Claude Code is included, and the September 2026 plan table lists both Opus and Sonnet on Pro. Reasonable usage limits for light to moderate coding. You get the full experience: codebase-aware editing, multi-file changes, command execution. Start here if you are evaluating Claude Code for the first time.

Max 5x ($100/mo): Five times Pro usage plus priority access at high-traffic times. The practical difference is how much Opus-tier work you can run before hitting limits. Complex refactors, architectural decisions, and multi-step autonomous workflows all benefit from the stronger model.

Max 20x ($200/mo): The highest individual tier, with 20x Pro usage. Usage limits still apply, but most daily users stop thinking about them. The sub-agent architecture, skills system, and autonomous loops are all included at every tier, but the $200 plan removes the friction of watching your usage. If you ship code every day, this plan pays for itself.

What you get at every tier: Full project context, multi-file editing, terminal command execution, MCP server integration, custom skills, memory system, parallel sub-agents. The difference between tiers is model quality and usage volume, not features.

Who it is for: Developers who want the strongest reasoning model applied to their entire codebase. Heavy users who run Claude Code for hours daily should go straight to $200 Max. Light users can start at $20 and upgrade when they hit limits.

More Claude Code reading: usage limits playbook, sub-agents, agent teams, and the setup guide.

Cursor#

Cursor is a VS Code fork with AI built into every interaction. Inline completions, a chat panel, multi-file Composer edits, and an agent mode that runs commands and iterates on results.

Free tier: Limited completions and a small number of premium model requests per month. Enough to evaluate the workflow. You get the editor, basic completions, and a taste of Composer. Not enough for daily development.

Pro ($20/mo): Cursor's current docs frame Pro as including $20 of agent usage at the underlying model API rates, plus bonus capacity beyond that guaranteed baseline. The workflow is still the same: unlimited tab completions, Composer, agent mode, Bugbot, and background agents. The important shift is budgeting. Your model choice now determines how quickly the included usage burns down.

Pro+ ($60/mo): Cursor now documents Pro+ as including $70 of agent usage plus bonus capacity. This is the practical tier for daily agent users who want better predictability than raw pay-as-you-go without immediately jumping to the highest tier.

Ultra ($200/mo): Cursor documents Ultra as including $400 of agent usage plus bonus capacity. That makes it the cleanest individual plan for multi-agent or automation-heavy users who want to stay inside one IDE surface.

Teams ($40/mo per active user): As of July 2026, Teams seat usage is split into two separate pools - one for Composer and Auto (first-party Cursor models) and one for third-party API models (Claude, GPT, Gemini) - plus admin dashboards, privacy controls, and spending limits. A Premium seat ($120/mo) offers 5x Standard usage for power users running heavy agent workloads. For organizations, that visibility matters as much as the sticker price.

Who it is for: Developers who prefer working inside an IDE with visual diffs and inline completions. The $20 Pro plan remains the best single-tool value in AI coding. For a detailed comparison with terminal-based tools, see Claude Code vs Cursor.

More Cursor reading: what Cursor is, Cursor AI code editor guide, Composer 2.0 deep dive, and Cursor vs Codex.

GitHub Copilot#

GitHub Copilot is the most widely adopted AI coding tool. It lives inside VS Code, JetBrains, and Neovim. The latest version includes agent mode with terminal access and multi-file editing.

Free tier: 2,000 code completions per month plus limited agent mode and CLI use, with no credit card required. Verified students get the separate Copilot Student plan.

Pro ($10/mo): Full completions, chat, and agent mode. Unlimited completions, $15 in monthly AI credits, model selection, and access to third-party agents (Claude Code and Codex) as of September 2026. Claude Sonnet 5.5 is now also available in Copilot, which changes the practical worker-model choice for routine bug fixes and reviews; see the Sonnet 5.5 release guide. The GitHub ecosystem integration is the differentiator. Copilot sees your issues, PRs, and CI results. When you reference a GitHub issue in a prompt, it pulls the full context automatically.

Business ($19/mo per seat): Everything in Pro, plus IP indemnity, organization-wide policy controls, audit logs, and the ability to exclude specific files from training. The IP protection alone makes this the default choice for companies with legal concerns about AI-generated code. As of June 1, 2026, GitHub's own billing docs treat higher-end usage as usage-based, so seat price is no longer the whole budget story. Read our GitHub Copilot guide.

Enterprise ($39/mo per seat): Adds fine-tuned models trained on your organization's codebase, knowledge bases, and web search inside the editor.

Watch out: GitHub still markets included premium-request allowances on the plan page, but the billing docs now spell out per-token AI-credit pricing for usage beyond the included allowance. If you lean on premium models or cloud agent runs, the overage path matters.

Who it is for: Teams already on GitHub. The ecosystem integration is unmatched. Individual developers get solid value at $10/mo, but the agent capabilities lag behind Cursor and Claude Code.

More Copilot reading: GitHub Copilot guide, Copilot coding agent CLI, Claude Sonnet 5.5 release guide, premium requests explained, and migrating from Copilot to Claude Code.

Windsurf#

Windsurf (formerly Codeium) has one of the most generous free tiers in the market. The Cascade agent chains multi-step operations together, and the editor handles TypeScript projects well.

Free tier: Generous autocomplete limits, access to Cascade agent mode, and a meaningful number of premium model requests. This is not a crippled trial. You can use Windsurf as your primary coding tool without paying for weeks or months. Tab completions are truly unlimited and cost no credits. For developers on a tight budget, this is the starting point.

Pro ($15/mo): About 1,000 prompts per month, faster response times, and priority access during peak usage. The $5 savings over Cursor Pro adds up to $60/year, and the free tier means you can evaluate thoroughly before committing.

Enterprise: Custom pricing with SSO, audit logs, and self-hosted deployment options.

Who it is for: Budget-conscious developers who want agent capabilities without paying $20/mo. The free tier is genuinely usable for real work. For head-to-heads with Cursor, see our Windsurf vs Cursor analysis and SWE-1 vs Cursor Composer comparison.

Cline#

Cline is the leading open-source AI coding extension for VS Code. It is free to install, runs with your own API keys, and supports both cloud models and local models through Ollama.

Free (BYOK): The extension itself is free. You bring your own API key for Claude, OpenAI, Gemini, Azure OpenAI, or other providers. Or you run local models through Ollama and pay nothing at all. There is no subscription, no credits system, no premium tier. You pay only what your chosen model provider charges.

Typical costs: Using Claude Sonnet through the Anthropic API, heavy Cline usage runs roughly $20-50/mo in API costs depending on context size and request volume. Using local models through Ollama, the cost is zero after hardware. Using OpenAI models, costs scale with the model tier and token usage.

What you get: Full agentic capabilities including multi-file editing, terminal command execution, MCP server integration, and iterative error correction. Cline reads your project, understands context, and operates autonomously on tasks. The workflow is comparable to Cursor's agent mode or Windsurf's Cascade, but without the bundled subscription.

Trade-offs: No built-in model optimization or caching. You pay raw API prices without the bulk discounts that Cursor or Copilot might negotiate. No vendor support - you troubleshoot via GitHub issues and community Discord. No built-in usage dashboards, so tracking costs requires external monitoring.

Who it is for: Developers who want full control over model choice and cost. Power users who would exceed subscription limits anyway. Teams that need to run local models for privacy or compliance. Anyone philosophically opposed to vendor lock-in.

More Cline reading: What is Cline, Aider vs Claude Code (another open-source option), and the AI coding tools comparison matrix.

OpenAI Codex#

OpenAI Codex brings OpenAI's coding models to the terminal and to cloud task execution. It follows the same CLI-agent pattern as Claude Code: read the project, reason about changes, execute them directly. For deeper context, read the OpenAI Codex guide, Codex vs Claude Code, and Claude Code vs Cursor vs Codex.

Free / Go / Plus: Codex access is now bundled across Free, Go, and Plus plans, according to OpenAI's Codex plan documentation. That makes Codex easier to trial than it was earlier this year, but the practical allowance still depends on plan-level agentic limits, task size, and whether you add credits.

ChatGPT Go ($8/mo): The budget tier with Codex access. Raised from $6/mo as of July 2026. Basic usage for light coding sessions.

ChatGPT Plus ($20/mo): Plus remains the lowest paid Codex entry point with the CLI and cloud execution mode. Be aware: developers report that the Plus tier can be exhausted quickly on heavy coding sessions. The larger context and higher usage budgets are reasons to evaluate Pro or business plans.

ChatGPT Pro (from $100/mo): Pro offers 5x or 20x Plus usage. As of September 28, 2026, OpenAI's help center says new sign-ups and upgrades to the $200 Pro plan are temporarily paused. Codex on paid plans now runs the GPT-6 family (Sol and Luna, with Astra listed in the usage table), and GPT-5.5 retires from Codex on October 14, 2026. Similar to Claude Max in positioning: designed for all-day usage. OpenAI's coding models have received positive reviews for TypeScript type inference and complex generic patterns.

ChatGPT Business: Codex is also included with ChatGPT Business. This is the missing middle tier for teams that need admin controls, workspace settings, and shared billing without jumping straight to a custom enterprise agreement. Business usage follows OpenAI's token-based Codex rate card, so check the official Codex rate card before modeling team cost.

Enterprise/Edu: Custom pricing with workspace features, admin controls, and dedicated capacity.

Who it is for: Developers already paying for ChatGPT Plus who want a terminal agent without an additional subscription. The cloud execution mode is a unique feature. For a direct comparison, see Cursor vs Codex.

More Codex reading: Codex Cloud security playbook, Codex managed agents on AWS, Codex security research preview, and GPT-5 Codex.

Augment#

Augment is a VS Code and JetBrains extension that focuses on large codebase understanding and structured planning. The Task List feature lets you review and edit a step-by-step plan before any code changes execute.

Dev plan (Free): Full access to Augment's core features including codebase indexing, chat, inline completions, and the Task List agent. Generous usage limits. Multi-model access including Claude and GPT-5.x. This is one of the most capable free tiers available because Augment is still in a growth phase and investing heavily in developer adoption.

Individual Pro ($50/mo): Higher usage limits, priority support, and access to additional models. The jump from free to $50 is steep, which means the free plan needs to be genuinely good to convert users. And it is.

Enterprise: Custom pricing with SSO, audit logs, and dedicated support. Team features for codebase-wide context sharing.

What makes it different: Augment indexes your entire codebase and maintains context across sessions in a way that most tools still struggle with. The Task List workflow gives you a review gate before the AI touches your code. For large monorepos and enterprise codebases, the context quality is a real differentiator.

Who it is for: Developers working on large codebases who want structured, reviewable AI assistance. The free tier makes it risk-free to try. If you work at a company with a 500K+ line codebase, Augment's context engine is worth evaluating. For the workflow angle, read the Augment Task List breakdown.

Gemini CLI#

Gemini CLI is Google's terminal-based coding agent. It connects to Gemini 2.5 Pro with one of the largest context windows available.

Free: The entire tool is free. No paid tier for normal usage. It connects to your Google account and uses the Gemini API's free tier, which advertises 1,000 requests per day. In practice, most of those requests hit Gemini Flash (the lighter model). Some developers report rate-limiting on Gemini Pro after just 4 to 15 large prompts.

Antigravity (free, or Google AI Pro $19.99/mo and Ultra from $99.99/mo): Google's agentic development platform has a free individual tier with weekly rate limits. Google AI Pro raises limits with a quota that refreshes every five hours, and Ultra adds the highest limits plus third-party models. As of September 2026, Ultra starts at $99.99/mo, which narrows the old gap between $20 and $250.

Google Cloud: For enterprise workloads, Gemini integrates with Vertex AI with its own token-based pricing. But for individual developers using the CLI, you pay nothing.

Who it is for: Every developer. There is no reason not to have it installed alongside your primary paid tool. Use it for high-volume tasks that do not justify burning premium credits: code review on large PRs, documentation generation, codebase analysis. See our Gemini CLI guide, best CLI tools for AI development, and Claude vs GPT coding comparison for broader model tradeoffs.

Zed#

Zed is a Rust-native code editor built for speed. Sub-50ms latency, GPU-accelerated rendering, and built-in AI features.

Editor (Free): The editor itself is free and open source. Fast, minimal, and designed for developers who care about performance.

Zed AI ($20/mo): Adds AI completions, inline editing, and chat powered by multiple models. The integration is tighter than bolt-on extensions because AI is built into the editor's core architecture.

Who it is for: Performance-focused developers who want a modern editor that is not a VS Code fork. The $20/mo AI tier competes directly with Cursor Pro on price but offers a fundamentally different editing experience.

Related editor comparisons: Cursor AI guide, Windsurf vs Cursor, and the AI coding tools comparison matrix.

Kiro#

Kiro is Amazon's AI coding tool that integrates with AWS services and uses a credit-based pricing system.

Free tier: Limited credits per month. Enough to evaluate the tool and run a few sessions.

Credit-based pricing: Different prompts cost different amounts depending on the model and complexity. Usage scales with your AWS billing. Developers have reported that credit consumption can be unpredictable, with bugs that drained limits unexpectedly during early access.

Who it is for: Teams deep in the AWS ecosystem who want AI coding integrated with their cloud workflow. The credit-based pricing makes it harder to predict monthly costs compared to flat-rate alternatives.

Devin#

Devin is the fully autonomous software engineer. You assign tasks through Slack or a web interface, and it works independently: setting up environments, writing code, running tests, opening PRs.

Beta ($20/mo): Individual access to Devin's autonomous capabilities. This pricing is significantly lower than earlier per-task pricing, reflecting Cognition's push for adoption.

Team ($500/mo per seat): Dedicated Devin capacity for organizations that want to parallelize work across AI agents and human developers.

Who it is for: Teams with strong test coverage who want to delegate well-defined tasks. The $500/seat team pricing is a major commitment, but if Devin handles even a few features per month autonomously, the math works.

Related autonomous-agent reading: what an AI coding agent is, agent infrastructure tools, long-running agent harnesses, and AI agent frameworks compared.

App Builders: v0, Lovable, Bolt#

These tools generate full applications from natural language descriptions. Different pricing model from coding assistants.

v0 (Vercel): Free tier with limited generations. Beyond that, a credit system where you pay per generation. Costs vary by complexity. Best for Next.js and shadcn/ui scaffolding.

Lovable: Free tier for one small project. Starter at $25/mo with more generations and the full template library. Best for MVPs and rapid product validation. See our Lovable app builder guide.

Bolt: Free tier with generous tokens. Pro at $25/mo for higher limits and faster generation. Best for browser-based prototyping with hands-on editing.

Who they are for: Developers validating product ideas quickly. Start with the free tiers and upgrade only when you have a specific project that benefits from rapid generation. For adjacent app-builder and UI workflows, read the Lovable app builder guide, create beautiful UI with Claude Code, and building SaaS with Claude Code.

Cost Per Feature Analysis#

Raw monthly cost does not tell the full story. Here is what each dollar actually buys across the tools that matter most for daily development.

Reasoning Quality Per Dollar#

The pattern: at the $200/mo tier, every major tool offers effectively unlimited usage. The real differentiation happens at $20/mo, where usage caps force you to choose which tasks deserve premium AI and which get handled by free tools.

Context Window Per Dollar#

Large codebase support varies dramatically by price tier.

Augment and Claude Code lead on context handling. Augment indexes your entire codebase and maintains that context across sessions. Claude Code loads your full project on every invocation. Codex context and credit behavior varies by ChatGPT plan and OpenAI's current rate card, which is why the Business tier matters for teams comparing Codex against seat-based tools.

Autonomy Per Dollar#

How much can each tool do without you babysitting it?

High autonomy (can run for minutes to hours unsupervised): Claude Code ($20+), Codex ($20+), Devin ($20+)

Medium autonomy (multi-step with checkpoints): Cursor Agent ($20+), Augment Task List (Free), Windsurf Cascade (Free), Cline (API costs)

Low autonomy (mostly reactive): Copilot ($10+), Zed AI ($20), basic completions on any tool

If autonomous operation is your priority, Claude Code at $200/mo offers the best ratio of capability to cost. You can run multi-hour coding sessions with sub-agent orchestration and skills-based workflows. No other tool matches that depth of autonomous operation at any price.

What I Actually Pay#

Here is the real cost of my daily development stack:

Claude Code handles the heavy lifting: autonomous refactors, multi-file changes, sub-agent orchestration, CI integration, and anything that benefits from deep reasoning. Cursor handles the fast iteration: UI tweaks, quick edits, visual diffs, and the work where IDE velocity matters more than raw intelligence. Augment's free tier fills a niche for large codebase navigation and structured planning. Gemini CLI handles high-volume code review and documentation at zero cost.

The $200 Max plan sounds expensive in isolation. In practice, it replaces hours of manual work every day. If you ship code professionally and your time is worth anything north of $50/hour, the math works within the first week.

Best Options by Budget#

$0/mo: The Free Stack#

Gemini CLI for terminal-based coding. Windsurf free tier for IDE work. Augment free tier for large codebase context. v0 free tier for UI generation.

This stack is genuinely usable. Gemini CLI's large context window handles codebase analysis. Windsurf's Cascade agent handles multi-step tasks. Augment adds structured planning with codebase-wide context. You can ship real projects without paying a cent. The tradeoff is reasoning quality on complex tasks and occasional rate limits during peak hours.

$10/mo: GitHub Ecosystem#

Copilot Pro at $10/mo. The cheapest paid option with real agent capabilities. Best if your workflow is GitHub-centric and you want completions, chat, and basic agent mode without spending more.

$20/mo: Best Single Tool#

Cursor Pro. The fastest AI coding environment for the money. Unlimited completions, full Composer, agent mode. If you can only pay for one tool, this is it.

Runner-up: Claude Code Pro at $20/mo if you prefer terminal-based workflows and stronger reasoning over IDE integration.

$40/mo: The Balanced Stack#

Cursor Pro ($20) plus Claude Pro ($20). Cursor for fast iteration and visual editing. Claude Code for autonomous tasks, refactors, and anything that benefits from stronger reasoning. This combination covers nearly every coding workflow.

$220/mo: The Power User Stack#

Claude Max ($200) plus Cursor Pro ($20). This is what heavy daily usage looks like. Claude Code runs for hours without usage anxiety. Cursor handles the quick edits and UI work. Add Augment (free) for codebase context and Gemini CLI (free) for high-volume tasks. This setup maximizes throughput for developers who ship code all day, every day.

$260/mo: Team Lead Stack#

Add Copilot Business ($19/seat) for GitHub ecosystem integration and IP indemnity. Add Devin ($20/mo beta) for delegating well-defined tasks. At this budget, you are optimizing for team productivity, compliance, and the ability to parallelize work across human developers and AI agents.

Hidden Costs to Watch#

Token overages on Cursor. Cursor Pro's usage limits depend heavily on which model you use. Higher-quality models burn through your allocation faster. Pro+ ($60/mo) and Ultra ($200/mo) exist specifically because Pro users kept hitting walls, and Cursor adjusted team pricing on June 1, 2026 to make spend more predictable for heavier seats. Know your usage patterns before assuming $20/mo is enough.

Codex credit and context math. OpenAI's current Codex docs say Codex is included with Plus, Pro, Business, and Enterprise/Edu plans, but the practical budget depends on credits, model usage, and context needs. Do not treat $20/mo Plus as equivalent to a full team coding budget. If you work on large codebases or run long tasks, compare Plus, Pro, and Business using the Codex rate card.

Copilot model multipliers. Using Opus-tier models on Copilot consumes 3x your premium request allocation. Your "unlimited" plan is effectively one-third as generous when you use the best models.

Gemini CLI rate limiting. The "1,000 requests per day" mostly hits the lighter Flash model. Real Gemini Pro access can throttle after 4-15 large prompts. Guarantee Pro access by bringing your own API key, which adds API costs on top.

Kiro credit unpredictability. Variable credit costs per prompt make monthly budgeting difficult. AWS has acknowledged bugs that drained credits unexpectedly. Budget a 2x buffer over your expected usage until the system stabilizes.

API keys for extended features. Some tools let you bring your own API keys for additional models. This sounds flexible until you realize you are paying the tool's subscription plus raw API costs. Check whether the features you need are included in the plan or require separate API billing.

Team seat math. A $20/mo tool becomes $2,400/year for a 10-person team. Copilot Business at $19/seat is $2,280/year for the same team. Enterprise plans with custom pricing often include volume discounts that individual pricing pages do not show. Always ask for team quotes before multiplying the per-seat price.

Context window limits on cheaper plans. Lower tiers often restrict the number of files you can reference in a single prompt. If you work on large TypeScript projects with hundreds of files, the difference between a plan that loads 50 files and one that loads 200 files directly affects output quality.

Model access varies by tier. Claude Pro now lists Opus access, but Fable runs on usage credits on Pro and is capped at 50% of weekly limits on Max, and heavy Opus use burns through Pro limits quickly. Cursor Pro gives you access to premium models, but the specific models available change as partnerships shift. Do not assume the model you tried during a free trial is the same one you get on the cheapest paid plan.

The Bottom Line#

The market has settled into clear tiers. Free tools (Gemini CLI, Windsurf, Augment Dev, v0) are good enough for real work. The $10-20/mo tier (Copilot Pro, Cursor Pro, Claude Pro, ChatGPT Plus with Codex) covers most developers. The $100-200/mo tier (Claude Max, Cursor Ultra, ChatGPT Pro with Codex) is for developers whose output directly generates revenue. Team plans like Copilot Business, Cursor Business, and ChatGPT Business sit between individual subscriptions and custom enterprise contracts.

Do not stack subscriptions you do not use daily. Pick one primary tool, add a free tier tool for overflow, and upgrade only when you consistently hit limits. The best pricing strategy is the one where every dollar spent maps to hours saved.

For recommendations on which tools to pick regardless of price, see our 10 best AI coding tools ranking. For head-to-head comparisons, check the tool comparison page, the agent comparison dashboard, the comparison hub, and the AI coding tools matrix.

Frequently Asked Questions#

How much does Claude Code cost?#

As of September 28, 2026, Claude Code is included in Claude Pro at $20/mo ($17/mo billed annually), Max 5x at $100/mo, and Max 20x at $200/mo. Team seats are $20 to $25 (Standard) or $100 to $125 (Premium), and Enterprise is $20 per seat plus usage at API rates. The Free plan does not include Claude Code. The tiers differ mainly in usage volume, not features.

Is Cursor free?#

Cursor has a limited free tier with basic completions and a small number of premium model requests per month. It is enough to evaluate the tool but not enough for daily development. The Pro plan at $20/mo is what most developers use for real work. Higher tiers at $60/mo (Pro+) and $200/mo (Ultra) are available for heavier usage.

How much does GitHub Copilot cost?#

As of September 28, 2026: Free is $0 with 2,000 completions a month, Pro is $10/mo with $15 in monthly AI credits, Pro+ is $39/mo with $70, and Max is $100/mo with $200. Business is $19 per seat and Enterprise $39 per seat. Copilot bills chat, agent mode, code review, and cloud agent work in AI credits, so treat the plan price as the entry point, not the full budget.

What is the best free AI coding tool?#

For terminal work, Gemini CLI. For IDE work, Windsurf's free tier. For large codebase context, Augment's Dev plan. Each excels at different tasks. Using all three together gives you a genuinely capable free stack that covers most coding workflows.

Is Augment Code free?#

Yes. Augment's Dev plan is free and includes codebase indexing, chat, inline completions, and the Task List agent with generous usage limits. The Individual Pro plan at $50/mo offers higher limits and priority support. The free tier is one of the most capable in the market because Augment is in a growth phase focused on developer adoption.

Cursor vs Claude Code: which is better value?#

At $20/mo, they serve different workflows. Cursor Pro is the best value for IDE-based development with visual diffs and inline completions. Claude Code Pro is the best value for terminal-based autonomous coding with stronger reasoning. Many developers use both: Cursor for fast iteration, Claude Code for complex tasks. See our full comparison.

What is the cheapest AI coding tool worth paying for?#

GitHub Copilot at $10/mo. You get completions, chat, and agent mode with GitHub ecosystem integration. If you need more capable agent features, Cursor Pro at $20/mo is the next step up and widely considered the best single-tool value at any price.

How much does Windsurf cost?#

Windsurf's free tier is available to all individual developers with generous limits including unlimited tab completions. The Pro plan costs $20/mo (raised from $15 in May 2026). A new Max tier at $200/mo offers higher quotas for power users, matching Cursor's tier structure. Enterprise pricing is custom. Windsurf has the strongest free tier among VS Code-fork editors.

Is OpenAI Codex free?#

Codex has limited access on ChatGPT Free and Go ($8/mo), using GPT-6 Luna in the desktop app subject to rollout. Plus ($20/mo) is the first tier with Codex across web, CLI, IDE, and iOS. Limits vary by model and task size, and OpenAI publishes estimated messages per five-hour window on its Codex pricing page. Always check OpenAI's Codex plan documentation and Codex rate card because credit rules changed in April 2026.

How much does Google Antigravity cost?#

Antigravity has a free individual tier with weekly rate limits and access to Gemini 3.8 Flash, Gemini 3.1 Pro, Claude Sonnet and Opus 4.6, and gpt-oss-120b. For higher limits you subscribe to Google AI Pro ($19.99/mo) or Google AI Ultra (from $99.99/mo), both of which can also buy AI credits for overage. Prices verified September 28, 2026.

Is Claude Cowork included in Claude Pro?#

Yes. Claude Cowork has no separate price; it works with paid Claude plans and runs on their usage. Team plans list Claude Code and Claude Cowork together. See Claude Cowork vs Claude Code for which to use when.

How much does Cursor Composer cost?#

Composer 2.5 is in Cursor's Cursor Models pool, which Pro ($20), Pro Plus ($60), and Ultra ($200) all include. Cursor lists its rate at $0.50 input and $2.50 output per million tokens, or $3/$15 in Fast mode, well below the third-party frontier models in the Other Models pool.

Do I need the $200/mo plan for any tool?#

Only if you code for 4+ hours daily and the tool is your primary development environment. The $200 tiers on Claude Code, Cursor, and Codex are designed for professional developers who would otherwise hit rate limits constantly. If you code a few hours per week, the $20 tier on any tool is more than enough.

Continue Reading#

• Where to Access AI Models in 2026 - one hub for every model's access routes, free tiers, and per-token prices
• AI Coding Tools Comparison Matrix 2026 - broad feature and workflow comparison across every major tool
• Claude Code vs Cursor vs Codex - the core three-tool decision broken down head-to-head
• Best AI Coding Tools 2026 - overall ranking after you understand the pricing landscape
• Frontier Model API Pricing June 2026 - API rate card for the models behind these coding tools
• Windsurf vs Cursor - how budget-friendly IDE agents compare with the category leader

Related apps#

• Agent Hub - One control panel for Claude Code, Codex, Gemini, Cursor, and 10+ AI coding harnesses. Desktop app for Mac.
• Cost Tape Cloud - Cloud cost tracking for AI agent runs. Cloud tier unlocks org-wide rollups, budgets, and alerts.

Related#

• Subscribe to DevDigest on YouTube for hands-on walkthroughs