--- format: "aidr-story-markdown/v1" id: "15a89a1e2ac8e4a26154a783aa443c69476cafd369b08e9fb1a653ed2b00e4de" canonical_url: "https://aidr.today/15a89a1e?lang=en" title: "xAI Launches 2.1 Trillion Parameter Grok 4.7 to Overtake GPT 5.6 Sol in Coding" lang: "en" requested_lang: "en" available_langs: ["en","vi"] translation_fallback: null fallback_fields: [] published_at: "2026-09-22T21:01:26.000Z" category: "Models" topics: ["xai","grok","coding","agent","reasoning","infra","benchmark","gpt"] source_urls: ["https://huggingnews.com/ai/xai-launches-21-trillion-parameter-grok-47-to-overtake-gpt-56-sol-in-cod-8e734762","https://x.com/CoinMarketCap/status/2102388744326111697","https://x.com/theo/status/2102268678779895993","https://x.com/theo/status/2102259624313549130","https://marketbrief.now/ai/xai-launches-21-trillion-parameter-grok-47-to-overtake-gpt-56-sol-in-cod-8e734762"] summary: "The latest artificial intelligence model from xAI focuses on agentic coding and professional knowledge work through its deployment on the Grok Build harness and API. Grok 4.7 utilizes a 2.1 trillion parameter architecture and achieved a score of 56 on the Artificial Analysis Coding Agent Index, ranking fourth behind Fable 5.1, GPT 6 Astra, and Opus 5. It recorded a 38% success rate on Terminal Bench 4.0, a jump from 20.3% for Grok 4.6, and a 19.6% score on the Harvey Legal Agent Benchmark. The model supports a 500,000 token context window with a price of $2 per 1 million input tokens and $6 per 1 million output tokens. Performance data from Artificial Analysis shows a decrease in efficiency, with the model using 81,000 output tokens per Intelligence Index task compared to 36,000 for Grok 4.6. Initial evaluations by the Vals benchmark suite showed a score of 54.2%, which improved to 10th place on the Vals Index after an SDK update." --- # xAI Launches 2\.1 Trillion Parameter Grok 4\.7 to Overtake GPT 5\.6 Sol in Coding > [Open the canonical story]() **Published:** 2026-09-22T21:01:26.000Z **Category:** Models **Topics:** xai, grok, coding, agent, reasoning, infra, benchmark, gpt ## Summary The latest artificial intelligence model from xAI focuses on agentic coding and professional knowledge work through its deployment on the Grok Build harness and API\. Grok 4\.7 utilizes a 2\.1 trillion parameter architecture and achieved a score of 56 on the Artificial Analysis Coding Agent Index, ranking fourth behind Fable 5\.1, GPT 6 Astra, and Opus 5\. It recorded a 38% success rate on Terminal Bench 4\.0, a jump from 20\.3% for Grok 4\.6, and a 19\.6% score on the Harvey Legal Agent Benchmark\. The model supports a 500,000 token context window with a price of $2 per 1 million input tokens and $6 per 1 million output tokens\. Performance data from Artificial Analysis shows a decrease in efficiency, with the model using 81,000 output tokens per Intelligence Index task compared to 36,000 for Grok 4\.6\. Initial evaluations by the Vals benchmark suite showed a score of 54\.2%, which improved to 10th place on the Vals Index after an SDK update\. ## Sources - [Story source]() - [Story source]() - [Supporting source]() - [Supporting source]() - [Story source]()