Grok 4.5 Ships: Opus-Class Intelligence at $2/$6 and a 500K Context — the Cheapest Flagship for Agents
xAI released Grok 4.5 on July 8, 2026: a 500K-token context window, API pricing of $2/$6 per 1M tokens, and an Artificial Analysis Intelligence Index of 54 (#8 of 188). For builders, the headline is cost — top-tier output at a fraction of Opus 4.8 and GPT-5.6 Sol.
Key facts: xAI released Grok 4.5 on July 8, 2026. The API model id is x-ai/grok-4.5. Context window is 500K tokens. Pricing is $2 per 1M input tokens and $6 per 1M output tokens (Source: OpenRouter). On Artificial Analysis’s Intelligence Index it scores 54 and ranks #8 of 188 models (Source: Artificial Analysis). It’s available through the xAI API, OpenRouter, Vercel, and Cursor.
What shipped
xAI calls Grok 4.5 its smartest model for coding, agentic tasks, and knowledge work. Two numbers define it for developers.
The context window is 500K tokens — roughly an entire mid-size codebase or a long research thread in a single prompt. (Source: OpenRouter model page.)
The intelligence score is 54 on the Artificial Analysis Intelligence Index, placing it #8 out of 188 tracked models — well above the ~30 average for comparable models. (Source: Artificial Analysis, artificialanalysis.ai/models/grok-4-5.)
The real story is price
Grok 4.5 lands at $2 input / $6 output per 1M tokens. Put that next to the other flagship-tier models and the gap is stark on the output side, where agentic workloads spend most of their tokens:
| Model | Input ($/1M) | Output ($/1M) | Context |
|---|---|---|---|
| Grok 4.5 | $2.00 | $6.00 | 500K |
| Claude Opus 4.8 | $5.00 | $25.00 | 1M |
| GPT-5.6 Sol | $5.00 | $30.00 | — |
Grok 4.5 pricing: OpenRouter. Opus 4.8: platform.claude.com. GPT-5.6 Sol: OpenAI official blog (see our GPT-5.6 switch guide).
That’s roughly one-fifth of Opus 4.8’s output cost and one-fifth of GPT-5.6 Sol’s — while sitting in the same top-8 intelligence band. For token-heavy agent loops, that ratio is the entire pitch.
What this means if you’re building agents
If you already route through a gateway, adopting Grok 4.5 is a one-line change: swap your model id to x-ai/grok-4.5 and re-run your evals. It’s live on the xAI API, OpenRouter, Vercel’s AI Gateway, and inside Cursor — the same IDE xAI’s parent SpaceX now owns (see SpaceX acquires Cursor/Anysphere).
Two caveats before you switch a production workload:
- Verify the benchmark claim on your own tasks. Third-party blogs have circulated a “#4, tops agentic tool use” ranking, but Artificial Analysis’s own page lists it at #8. Rankings shift as the index updates — run your real eval set, not someone’s leaderboard screenshot.
- Context is 500K, not 1M. If your pipeline assumes a full million-token window (as Opus 4.8 offers), Grok 4.5 will truncate. Check your longest prompts.
For a side-by-side on switching your flagship model cleanly — model IDs, cost math, and migration gotchas — see our guide on how to switch to Claude Opus 4.8, the pattern applies to any flagship swap.
Sources
Source: xAI Official