Codex vs Claude Code in July 2026: Opus 5, GPT-5.6, and the Post-Fable-5 Landscape

Official Sources#

Last updated: July 30, 2026

What Changed on July 30, 2026#

This post was originally published June 10 and refreshed June 15, 2026. Since then, the landscape shifted across every dimension of the comparison:

• Opus 5 launched July 24 at $5/$25 per MTok - near-Fable-5 intelligence at half the price, replacing Opus 4.8 as the Claude Code workhorse. Fable 5 remains available via API ($10/$50) but is no longer included in subscription plans after going API-only July 9.
• GPT-5.6 family went GA - GPT-5.6 Sol replaced GPT-5.5 as Codex's flagship model, with Terra and Luna tiers below it. The GPT-5.5, GPT-5.4, and GPT-5.4-mini models remain available but are now previous-generation.
• Codex hit 8 million users as of July 15, up 7x since February 2026, driven by the GPT-5.6 launch and ChatGPT Work consolidation.
• Claude Code Max plans restructured: $100/month (5x usage) with Opus 5 included, $200 for expanded capacity. Fable 5 access requires usage credits at API rates.
• Sonnet 5 introductory pricing ($2/$10 per MTok) confirmed through August 31, 2026, making it the cheapest frontier-capable model on the Claude side.
• Claude Code 2.1.200+ shipped auto-mode for unattended task completion, agent teams maturation, and the Agent SDK reaching production stability.

The model situation has flipped (again)#

In June, the story was Fable 5's ceiling versus the June 22 subscription cliff. Both are now resolved in opposite directions.

Claude Code side. Opus 5 launched July 24 at $5 input / $25 output per million tokens - the same pricing as Opus 4.8 but delivering near-Fable-5 capability. On CursorBench and internal Anthropic evaluations, Opus 5 scores within striking distance of Fable 5 on long-horizon coding tasks while costing half as much. For developers on Max plans ($100-$200/month), Opus 5 is the default workhorse model with no additional API charges.

Fable 5 went API-only on July 9 after a series of deadline extensions. It remains available at $10/$50 per MTok via the API for teams that want the absolute ceiling, but it is no longer included in any subscription plan. The June 22 cliff that this post's original version discussed came and went with a last-minute extension, followed by the final API-only switch on July 9. See Fable 5 pricing cost per task for the full timeline and cost modeling.

Codex side. The GPT-5.6 family went GA on July 9, replacing GPT-5.5 as the recommended default. Codex now ships three tiers: GPT-5.6 Sol (flagship, replaces GPT-5.5), GPT-5.6 Terra (balanced everyday workhorse), and GPT-5.6 Luna (fast and affordable for clear tasks). The older GPT-5.5, GPT-5.4, and GPT-5.4-mini models remain available but are designated as previous-generation. Codex also supports Ultra mode, which uses subagents to parallelize complex work across multiple model instances, and Max reasoning for deeper single-task analysis.

The headline in July 2026: Claude Code has Opus 5 at a price that changes the value equation. Codex has GPT-5.6 Sol bundled into ChatGPT plans (Plus at $20/month, Pro at $200/month) with no usage-credit cliff. The model quality gap that favored Claude in June has narrowed - Opus 5 is strong at $5/$25, but GPT-5.6 Sol is competitive on SWE-bench and Terminal-Bench while being included in flat-rate subscriptions.

Agentic surfaces: each tool grew in different directions#

Claude Code deepened its orchestration stack. The core primitives remain: skills, hooks, subagents (nesting five levels deep), routines (cloud-managed schedules), channels, and the Agent SDK. New since June: auto-mode lets Claude Code run unattended for approved task types, agent teams reached production stability for multi-agent parallel work, and the Agent SDK (now at production v1.0) lets teams build fully custom orchestration pipelines. The surface matrix - CLI, VS Code, JetBrains, Desktop app, Web, iOS/Android remote control - all share the same CLAUDE.md files, settings, and MCP servers.

Codex continues to invest in environmental access. GPT-5.6 Sol brought Ultra mode (parallel subagents for complex work) and Max reasoning (deeper single-task depth). Codex's automations, computer use (macOS + Windows, outside EEA), in-app browser, Chrome extension, worktrees, and Codex SDK are all stable and documented. The ChatGPT Work consolidation merged Codex's and ChatGPT's desktop experience. Codex hit 8 million users as of July 15, with the GPT-5.6 launch and Work features driving adoption. The Codex Security plugin is now in beta for enterprise teams.

The practical difference from June holds: Claude Code's agentic model is agent-to-agent coordination with persistent cloud infrastructure. Codex's is task-to-environment with strong local desktop control. But the gap narrowed - Claude Code added auto-mode for unattended work (answering a Codex strength), while Codex added Ultra subagent parallelism (answering a Claude Code strength).

Headless and CI#

Both tools remain genuinely usable in pipelines.

Claude Code ships claude -p for non-interactive execution, GitHub Actions integration, GitLab CI/CD support, and the Agent SDK for custom orchestration. The GitHub Code Review feature provides turnkey PR automation. New since June: improved claude -p reliability in CI pipelines and better error reporting for headless runs.

Codex exposes codex exec for headless runs, the Codex SDK, a GitHub Action, and non-interactive mode. The Codex App Server lets enterprises self-host their agent infrastructure. Codex also added the Codex Security plugin for CI pipeline security scanning.

Both tools support AGENTS.md / CLAUDE.md context files for CI - no meaningful difference.

Ecosystem: MCP vs Codex SDK plus plugins#

Claude Code remains at the center of the MCP ecosystem. MCP is an open standard for connecting tools to external data - Google Drive, Jira, Slack, custom tooling - and the same server configuration works across all Claude Code surfaces. The MCP 2026-07-28 final spec shipped MCP Apps (interactive sandboxed UIs for MCP servers) as an official extension, expanding the ecosystem significantly.

Codex supports MCP connectors as well, plus its own plugin system, skills system, and Chronicle memory system. The Codex SDK lets you build programmatic integrations. Amazon Bedrock deployment is available for enterprise teams.

The ecosystem gap has widened since June: MCP's July 28 final spec with MCP Apps gives Claude Code a structural advantage for tool integrations that work everywhere. Codex's plugin architecture remains proprietary but well-documented.

What developers actually say#

The original HN thread this post cited - "Is Codex really on par with Claude Code?" - has aged well. The developer who switched to Codex cited its sandbox concept and ability to complete tasks without intervention. The developer who stayed with Claude Code valued persistent sessions, background tasks, and the hook/skill system.

Three months later, the community consensus is more polarized. Codex's 8M user milestone and GPT-5.6 launch have attracted a wave of new users, while Claude Code's Opus 5 launch and auto-mode have addressed the two biggest complaints: model pricing uncertainty and the need to manually approve every action.

Head-to-head comparison table#

Cost per persona after the Fable 5 transition#

The cost landscape is clearer now than in June.

Individual developer on a flat subscription. If you are on Claude Max ($100/month), Opus 5 is included with no extra charge. Fable 5 requires usage credits at API rates ($10/$50 per MTok). For $20/month, Codex on ChatGPT Plus gives you GPT-5.6 Sol with no similar cliff. For budget-conscious solo devs, Codex remains the more predictable value at the entry tier. At the Max tier, Opus 5's inclusion makes Claude Code compelling for heavy users.

API/power user. Opus 5 at $5/$25 is the strongest price-performance play in the market - near-Fable-5 quality at less than half the cost. GPT-5.6 Sol at $5/$30 is close on benchmarks. For production agents, Opus 5 is the recommended default, with Fable 5 available as a premium escalation path.

Enterprise team using CI heavily. Claude Code's routines run on Anthropic infrastructure. Codex automations run locally or via App Server. Claude Code's GitHub Code Review is turnkey; Codex's App Server offers self-managed deployment. Both solid, different operational models.

Team that needs GUI automation. Codex has computer use today. Claude Code does not. This remains a clear differentiator.

Who should pick which tool right now#

Pick Claude Code if: You want Opus 5 at the best price-performance ratio on the market. You need cloud-managed scheduled agents via routines. You are invested in the open MCP ecosystem. You run multi-agent coordination where a lead agent orchestrates parallel workers. You value the ability to move sessions between terminal, IDE, desktop, web, and mobile seamlessly.

Pick Codex if: You are on a ChatGPT plan and want GPT-5.6 Sol included with no usage-credit complexity. You need GUI automation, computer use, or visual browser control. You want a single subscription covering both chat and code. You need the Codex App Server for self-hosted enterprise deployment. You want the Codex Security plugin.

Run both if: Most teams with serious agentic workflows end up here. Codex for environmental tasks and GUI automation; Claude Code for complex orchestration and Opus 5's price-performance combination. The AGENTS.md and CLAUDE.md conventions are close enough that project context stays in sync.

FAQ#

Is Fable 5 still available in Claude Code?#

Fable 5 went API-only on July 9, 2026. It is not included in any subscription plan. Access requires usage credits at $10/$50 per million tokens via the Claude API. Opus 5 ($5/$25) is the default workhorse model on Max plans and is the recommended daily driver for most teams.

What model does Codex use by default in July 2026?#

GPT-5.6 Sol is the recommended default for Codex as of July 2026, per the Codex models page. GPT-5.6 Terra is the balanced everyday option, and GPT-5.6 Luna is the fast, affordable tier. GPT-5.5, GPT-5.4, and GPT-5.4-mini remain available as previous-generation models.

Does Codex have computer use?#

Yes, Codex has computer use on macOS and Windows (outside the European Economic Area, UK, and Switzerland). Documented at developers.openai.com/codex/app/computer-use. Claude Code does not offer equivalent GUI automation.

Which tool is better for CI/CD pipelines?#

Both are production-ready. Claude Code has turnkey GitHub Code Review and cloud-managed routines that run without a local machine. Codex has codex exec, a GitHub Action, and the App Server for self-hosted deployments. The choice depends on whether you want Anthropic-managed execution (Claude Code routines) or self-hosted infrastructure (Codex App Server).

How does Opus 5 pricing compare to GPT-5.6 Sol?#

Opus 5: $5 input / $25 output per MTok. GPT-5.6 Sol: $5 input / $30 output per MTok. Both are nearly identical in price. Opus 5 has a slight edge on long-horizon coding benchmarks at this price point, making it the strongest price-performance play in the market as of July 2026.

Continue Reading#

• Claude Opus 5 vs Opus 4.8 vs Fable 5 Comparison - Full benchmark breakdown and decision guide
• AI Coding Tools Pricing 2026 - Current pricing for every major tool and model
• Frontier Model API Pricing - API rate comparison across providers
• Claude Code vs Cursor 2026 - The other major Claude Code comparison
• Aider vs Claude Code 2026 - Open-source alternative to Claude Code
• Migrating from Windsurf to Claude Code: The Practical 2026 Guide

Sources#

• Anthropic announcement: Claude Opus 5 - July 24, 2026
• Anthropic announcement: Claude Fable 5 and Claude Mythos 5 - June 9, 2026
• OpenAI: Codex developer documentation - accessed July 30, 2026
• OpenAI: Codex models page - accessed July 30, 2026 (GPT-5.6 Sol/Terra/Luna)
• OpenAI: Codex automations documentation - accessed July 30, 2026
• OpenAI: Codex computer use documentation - accessed July 30, 2026
• Anthropic: Claude Code documentation - accessed July 30, 2026
• Anthropic: Claude Code releases - v2.1.200+
• OpenAI: Codex releases - accessed July 30, 2026
• Claude pricing: claude.com/pricing - Opus 5 $5/$25, Fable 5 $10/$50, Sonnet 5 promo $2/$10 - accessed July 30, 2026
• OpenAI pricing: openai.com/api/pricing/ - GPT-5.6 Sol $5/$30 - accessed July 30, 2026
• CursorBench: cursor.com/benchmarks - accessed July 30, 2026
• HN discussion: Is Codex really on par with Claude Code? - developer quotes