Overview
Grok 4.6 launched on August 12, 2026 as xAI's new frontier model for coding, agentic tasks, and knowledge work. It builds on Grok 4.5 with a stronger focus on long-running agents, multi-step research, codebase work, interactive project building, and visual application generation. The model is available through the xAI API, Grok Build, Cursor, and partner gateways including OpenRouter, Vercel, and Cloudflare.
The headline tradeoff is continuity rather than a context expansion: Grok 4.6 keeps a 500K-token context window and $2 input / $6 output base pricing, but adds xhigh reasoning, stronger vendor-published benchmark results, and better behavior on extended agent trajectories. xAI positions it as matching GPT-5.6 Sol on the Artificial Analysis Intelligence Index while still competing closely with Anthropic's Fable 5 across agentic coding and knowledge-work benchmarks.
What's New
Better Long-Running Agents
xAI describes Grok 4.6 as focused on staying with complex tasks across many steps: researching unfamiliar topics, analyzing information, working through a codebase, and turning product ideas into working applications or artifacts. Compared with Grok 4.5, xAI says it saw stronger first passes on visual and interactive projects, plus more self-testing and verification on longer trajectories.
xhigh Reasoning Effort
The API model page lists reasoning modes as low, medium, high by default, and xhigh. That gives builders another quality-latency control for difficult agentic tasks. For production workflows, the practical migration point is to test Grok 4.6 under the same reasoning effort and prompt-cache configuration as Grok 4.5 before comparing cost or latency.
Stronger Vendor-Published Benchmarks
xAI's announcement reports Grok 4.6 High at 61 on the AA Intelligence Index versus 56 for Grok 4.5 High, 65.9% on DeepSWE v1.1 versus 54%, 61.3% on FrontierCode v1.1 Extended versus 56.6%, and 57.5% on APEX-Agents versus 47.1%. It also reports GDPVal-AA v2 at 1753 versus 1526 and AA-Briefcase at 1577 versus 1313. These are vendor-published figures, so teams should treat them as directional and validate on their own workloads.
Broader Launch-Day Access
Grok 4.6 is available through the xAI API, Grok Build, Cursor, OpenRouter, Vercel, and Cloudflare. xAI also says Grok Build and Cursor include 2x usage for the first week after launch, making those paths useful for evaluation before committing significant API spend.
Performance Benchmarks
Vendor-published results from xAI's launch post:
| Benchmark | Grok 4.6 High | Grok 4.5 High | GPT-5.6 Sol Max | Fable 5 Max |
|---|---|---|---|---|
| AA Intelligence Index | 61 | 56 | 61 | 62 |
| GDPVal-AA v2 | 1753 | 1526 | 1728 | 1741 |
| CursorBench v3.2 | 69.9% | 66.7% | 67.2% | 70.5% |
| DeepSWE v1.1 | 65.9% | 54% | 73% | 70% |
| FrontierCode v1.1 Extended | 61.3% | 56.6% | 60.6% | 63.6% |
| APEX-Agents | 57.5% | 47.1% | 56.7% | 59.2% |
| Terminal-Bench v3.0 | 26% | 15.7% | 34.6% | 34.1% |
| APEX-SWE | 56.4% | 53.6% | - | 58.8% |
| AA-Briefcase | 1577 | 1313 | 1502 | 1574 |
| Harvey LAB (Vals) | 15.8% | 12.9% | 2.5% | 11.3% |
The most useful comparison for Grok users is Grok 4.6 versus Grok 4.5: every published row improves. Against competitors, the picture is mixed. Grok 4.6 matches GPT-5.6 Sol on AA Intelligence, beats it on several knowledge-work rows, but trails on DeepSWE and Terminal-Bench. Fable 5 still leads many coding-heavy rows.
Migration Guide
Teams moving from Grok 4.5 should start with the model ID: use grok-4.6 in xAI API calls. The context window remains 500K tokens, so prompts that already fit Grok 4.5 should not need context restructuring. Pricing also starts at the same $2 input and $6 output per 1M tokens, but cached input is now listed at $0.50 per 1M tokens and long-context rates apply once prompts reach 200K tokens.
For cost-sensitive agents, set a prompt cache key so repeated context can hit cache reliably. For quality-sensitive agents, compare high and xhigh reasoning on a fixed evaluation set before switching defaults. For Cursor or Grok Build users, the first-week 2x included usage is a low-friction way to test the model on real coding and project-building tasks.
Pricing & Plans
| Access Path | Price | Notes |
|---|---|---|
| xAI API, prompts below 200K tokens | $2 input / $0.50 cached input / $6 output per 1M tokens | Standard Grok 4.6 pricing |
| xAI API, prompts at or above 200K tokens | $4 input / $1 cached input / $12 output per 1M tokens | Long-context pricing applies to the request |
| Grok Build | Included usage varies | xAI offered 2x included usage for the first launch week |
| Cursor | Included usage varies by plan | Available in Cursor on launch day |
| Partner gateways | Varies by provider | xAI lists OpenRouter, Vercel, and Cloudflare |
The model page lists a 500K context window, text and image input, text output, no text output limit, and support for Responses API and Chat Completions. Server-side tools such as web search, X search, and code execution add separate tool-invocation costs when used.
Best For
- Engineering teams already using Cursor or Grok Build for coding agents and project generation
- Developers who liked Grok 4.5's cost profile but need stronger long-running agent behavior
- Product teams turning broad application ideas into first-pass working prototypes
- Research and analysis workflows that need a model to sustain multi-step work across tools
- Teams comparing lower-cost Grok API pricing against OpenAI and Anthropic frontier models
FAQ
How is Grok 4.6 different from Grok 4.5?
Grok 4.6 keeps the same 500K context size and headline $2/$6 API pricing, but improves long-running agent behavior, adds xhigh reasoning, and posts stronger vendor-published benchmark results across coding, agentic, and knowledge-work evaluations.
What is the API model name?
Use grok-4.6. xAI documents it for the Responses API and Chat Completions, with text and image input and text output.
How much does Grok 4.6 cost?
Below 200K prompt tokens, xAI lists $2 per 1M input tokens, $0.50 per 1M cached input tokens, and $6 per 1M output tokens. At or above 200K prompt tokens, rates are $4, $1, and $12 respectively.
Is Grok 4.6 better than Fable 5 or GPT-5.6 Sol?
It depends on the benchmark and workload. xAI reports Grok 4.6 matching GPT-5.6 Sol on AA Intelligence and beating it on several knowledge-work rows, while Fable 5 still leads many coding-heavy rows. Treat those numbers as starting points for internal testing.
Where can I use Grok 4.6?
xAI lists Grok 4.6 as available in the xAI API, Grok Build, Cursor, OpenRouter, Vercel, and Cloudflare. Availability and limits may vary by account, plan, and provider.
