
TL;DR
xAI launched Grok 4.5, trained on trillions of Cursor interaction tokens. At $2/M input pricing, it undercuts Claude and GPT while benchmarking near Opus 4.7 level.
xAI released Grok 4.5 yesterday - their first model trained on Cursor's massive developer interaction dataset. At $2/M input tokens (versus $5/M for Opus 4.8 and GPT 5.5), it's the cheapest frontier-tier coding model on the market. The Hacker News thread hit 672 points and 1,077 comments, with debate centering on whether the pricing is sustainable, whether developers will adopt it despite xAI's reputation, and how it stacks against established players.
The key differentiator is the training data. After xAI acquired Cursor earlier this year, they gained access to trillions of tokens of developer-AI interaction data - not just code, but the full workflow of how developers prompt, iterate, and refine with AI assistance.
From the Cursor blog:
Grok 4.5 represents Cursor's "most intelligent model and the first we've built for more than software engineering." It handles complex, long-running tasks across software engineering, data science, finance, legal work, and general computer-based problem-solving.
| Model | Input | Output |
|---|---|---|
| Grok 4.5 | $2/M | $6/M |
| Grok 4.5 Fast | $4/M | $18/M |
| GPT 5.5 | $5/M | $30/M |
| Claude Opus 4.8 | $5/M | $25/M |
| Claude Fable 5 | $10/M | $50/M |
Note: Grok 4.5 pricing doubles to $4/$12 for contexts over 200K tokens.
xAI claims Grok 4.5 performs at "around Opus 4.7 level" - roughly a generation behind current frontier but at 40% of the price. There's a caveat in the fine print: "Grok 4.5 has an advantage on CursorBench because an earlier snapshot of the Cursor codebase was accidentally included in training."
The model uses a mixture-of-experts architecture with 500K context (200K at base pricing).
The thread generated intense discussion across several themes.
Many commenters said they won't use Grok regardless of quality:
"Even without the politics, Elon has shown that he will weaponize his platforms against people/companies he personally doesn't like. Using Grok is therefore a supply chain risk and it's not nearly good enough to offset that risk."
Others pushed back on making technical decisions based on politics:
"Americans are 4% of the world's population, and even among those 4% at least half don't give a shit. The rest of us give even less of a shit, we don't have the luxury to be principled."
The aggressive pricing raised questions about sustainability:
"Why would having more costs and less income allow them to pass savings on to the end user?"
Some theorized xAI has excess compute capacity from their massive GPU build-out that's sitting partially idle, letting them price aggressively to gain market share. Others noted xAI reported $2.5B in operating losses last quarter.
The Grok Build CLI is now available for SuperGrok subscribers ($300/year), competing directly with Claude Code and Codex. Early users report it's "the fastest I've used in terms of responsiveness" but the model quality lagged until now.
With Grok 4.5 available in Cursor's harness, xAI finally has competitive infrastructure for agentic coding workflows.
Several commenters noted that Cursor's existing Composer 2.5 model - which is much cheaper to run - handles most coding tasks well:
"Composer 2.5 is so underrated IMO. I built a really feature rich application, insanely complicated, close to 200k LOC since it came out and for the most part it ran like a champ."
The question is whether Grok 4.5's broader training makes it worth the cost premium over task-specific models.
Newsletter
Get the weekly deep dive
Tutorials on Claude Code, AI agents, and dev tools, delivered free every week.
From the archive
Jul 8, 2026 • 7 min read
Jul 8, 2026 • 6 min read
Jul 7, 2026 • 6 min read
Jul 7, 2026 • 5 min read
Three things to watch:
1. Cursor data is a competitive moat. Training on real developer workflows - not just code, but the iterative prompting patterns of millions of users - produces models that understand how developers actually work. This is data that OpenAI and Anthropic don't have at this scale.
2. The price war continues. At $2/M input, Grok 4.5 undercuts every comparable model. If xAI can sustain this pricing (a big if given their burn rate), it puts pressure on Anthropic and OpenAI to respond.
3. Model routing gets more interesting. Many teams already route between models based on task complexity. A cheap, fast model for simple completions; an expensive reasoning model for complex tasks. Grok 4.5 slots into this matrix as "frontier-ish at Sonnet prices."
xAI's strategy is becoming clearer: use Cursor's distribution to capture developer workflows, train models on that data, and price aggressively to gain share. The Colossus 2 datacenter is training 5T and 10T parameter models that could extend this lead.
For developers evaluating Grok 4.5:
The model is available now via xAI's API and through Cursor's desktop, web, and CLI interfaces.
$2 per million input tokens, $6 per million output tokens for contexts under 200K. Pricing doubles for larger contexts.
xAI claims Grok 4.5 performs at "Opus 4.7 level" - roughly one generation behind Opus 4.8 - but at 40% of the cost.
Yes. Grok 4.5 is available in Cursor's desktop, web, iOS, and CLI interfaces for subscribers.
Training data. Grok 4.5 was trained on trillions of tokens from Cursor's user interaction dataset, capturing real developer-AI workflows rather than just static code.
Read next
xAI has launched Grok 4, claiming the title of the world's most powerful AI model. With a $300/month Super Grok tier, saturated AMI benchmarks, and a coding model on the horizon, this is xAI's bigge...
7 min readGrok Build is xAI's agentic CLI with 8 parallel subagents, a plan-first workflow, and Arena Mode for competing outputs. Installation, pricing, real commands, and how it compares to Claude Code and Codex.
9 min readCursor is a VS Code fork with AI at the center instead of bolted on. What it actually does, how it compares to Copilot and Claude Code, and when to reach for it - every fact checked against the official docs.
11 min readTechnical content at the intersection of AI and development. Building with AI agents, Claude Code, and modern dev tools - then showing you exactly how it works.
xAI's model with real-time X/Twitter data access. Grok 3 rivals top models on reasoning. Built-in web search and current...
View ToolHigh-performance code editor built in Rust with native AI integration. Sub-millisecond input latency. Built-in assistant...
View ToolAnthropic's smallest Claude 4.5 model. Near-frontier coding performance at one-third the cost of Sonnet 4 and up to 4-5x...
View ToolFactory AI's terminal coding agent. Runs Anthropic and OpenAI models in one subscription. Handles full tasks end-to-end...
View ToolInstall Ollama and LM Studio, pull your first model, and run AI locally for coding, chat, and automation - with zero cloud dependency.
Getting StartedReal-time prompt loop with history, completions, and multiline input.
Claude CodeHold-to-record voice input on macOS, Linux, and Windows.
Claude Code
In this video, I introduce the beta release of Grok 2 and Grok 2 Mini. I discuss the new models available on the X platform and their impressive performance, including their ranking on the...

Learn The Fundamentals Of Becoming An AI Engineer On Scrimba; https://scrimba.com/the-ai-engineer-path-c02v?via=developersdigest In this video, I dive into the key highlights of the groundbreaking...

Mercury Two: The First Reasoning Diffusion LLM (1,000+ tokens/sec) - Speed Without Sacrificing Quality Inception Labs releases Mercury Two, a reasoning diffusion-based LLM that exceeds 1,000 tokens p

xAI has launched Grok 4, claiming the title of the world's most powerful AI model. With a $300/month Super Grok tier, sa...

Grok Build is xAI's agentic CLI with 8 parallel subagents, a plan-first workflow, and Arena Mode for competing outputs....

Cursor is a VS Code fork with AI at the center instead of bolted on. What it actually does, how it compares to Copilot a...

Fable 5 landed on June 9, GitHub Copilot rewired its billing on June 1, and the tool-stack decisions you made in Q1 may...

The Bun runtime completed an AI-assisted rewrite from Zig to Rust, fixing memory safety issues and improving performance...

Martin Alderson's argument for why open-weights models like GLM 5.2 will compress frontier lab margins is sparking debat...

New tutorials, open-source projects, and deep dives on coding agents - delivered weekly.