xAI Launches Grok 4.6, Claims Parity with GPT-5.6 Sol and Near-Parity with Claude Fable 5

xAI released Grok 4.6 today, claiming the model matches OpenAI's GPT-5.6 Sol and comes within one point of Anthropic's Claude Fable 5 Max on a standard intelligence benchmark.

Benchmark claims

According to xAI, Grok 4.6 scored 61 on the Artificial Analysis Intelligence Index. That figure ties GPT-5.6 Sol at its maximum reasoning setting and sits one point behind Claude Fable 5 Max. xAI also reports improvements over the previous Grok 4.5 release across every evaluation it listed, including CursorBench, FrontierCode, APEX-Agents, and Terminal-Bench — though the company did not publish the underlying scores for those benchmarks in its announcement.

These are xAI's own reported figures. None of the comparative claims against GPT-5.6 Sol or Claude Fable 5 Max have been independently verified by a third party.

What xAI says changed

xAI attributes the gains to a longer supplemental training run, expanded reinforcement learning focused on coding and knowledge work, and stronger engineering data in the training mix. The company positions Grok 4.6 for long-running agent tasks, unfamiliar codebase navigation, and what it describes as increased self-checking of its own output before completing a task. No parameter count, context window size, or training data cutoff date was disclosed.

Pricing and availability

Grok 4.6 is live now through xAI's API, Grok Build, Cursor, OpenRouter, Vercel, and Cloudflare. Standard API pricing is $2 per million input tokens and $6 per million output tokens. A faster variant is priced at double that rate — roughly $4 per million input tokens and $12 per million output tokens, based on xAI's stated multiplier. Cursor and Grok Build users get double the included Grok 4.6 usage for the first week following launch.

Context

The release follows xAI's acquisition of Cursor in June 2026 and the launch of the Grok Bot agent for iPhone and Mac earlier this week. Together, the moves suggest xAI is building out a full agentic coding stack — model, IDE, and standalone agent — rather than competing on model quality alone.

What this means

Grok 4.6's headline number puts it in the same tier as GPT-5.6 Sol and just behind Claude Fable 5 Max, at least by xAI's own measurement. Pricing at $2/$6 per million tokens undercuts what OpenAI and Anthropic typically charge for comparable frontier tiers, which is likely the more consequential detail for developers choosing between models on cost rather than marginal intelligence differences. The real test will be how Grok 4.6 performs on independently run coding and agent benchmarks now that it's live across Cursor, OpenRouter, and other third-party platforms — none of xAI's comparative claims have been verified outside the company's own announcement.