The model tied GPT-5.6 Sol and finished one point behind Claude Fable 5 on a composite performance index, while carrying lower standard API prices than both.
xAI released Grok 4.6, a model built to continue working through longer, multi-step assignments, and priced it below comparable models from OpenAI and Anthropic.
The company is offering Grok 4.6 through its application programming interface (API), which lets businesses connect the model to their own software. xAI says the model can research topics, analyze information, work across software code, and build applications over many steps.
Benchmark results show mixed competition
xAI reported that Grok 4.6 scored 61 on the Artificial Analysis Intelligence Index, which combines nine performance tests. That tied OpenAI’s GPT-5.6 Sol and placed Grok 4.6 one point behind Anthropic’s Claude Fable 5.
Results varied across the individual tests. Grok 4.6 led both competing models on tests of professional work and legal tasks but trailed them on software-engineering and command-line tasks.
Higher scores indicate better performance. xAI said the competitor figures came from the developers’ published system cards or public test leaderboards. The comparison therefore combines xAI’s results with figures published by other organizations.
Grok 4.6 pricing undercuts two leading models
For standard API use with prompts below 200,000 tokens, Grok 4.6 costs less per token than GPT-5.6 Sol and Claude Fable 5. Tokens are pieces of text that AI companies use to calculate model usage.
At those rates, Grok 4.6 input costs are 60% below GPT-5.6 Sol and 80% below Claude Fable 5. Its output costs are 80% below GPT-5.6 Sol and 88% below Claude Fable 5.
The comparison covers published per-token prices only, not the total cost to complete a given task. Models may use different numbers of tokens to produce comparable work. xAI and OpenAI charge higher rates for prompts above 200,000 tokens, while Anthropic applies Claude Fable 5’s standard rate across its full context limit.
xAI provides few safety-testing details
xAI said it improved Grok 4.6’s safeguards and conducted its broadest set of capability and safeguard tests before releasing the model. The company also referred to extensive testing after release and by outside organizations.
The announcement does not identify what risks the tests uncovered, name the outside organizations, provide detailed test results, or explain what safeguards changed. It says the protections cover legitimate uses such as repairing software vulnerabilities, engineering work, and AI research.
Grok 4.6 is available through several services
Grok 4.6 is available through xAI’s API, Cursor, Grok Build, OpenRouter, Vercel, and Cloudflare. Its standard API price starts at $2 per million input tokens and $6 per million output tokens. A faster version costs twice those rates.
xAI did not say whether or when it would publish more information about the model’s safety tests and safeguards.
