Grok 4.7
A minor iterative release within xAI's Grok 4 model family, with optimizations in multi-turn conversational coherence, creative expression, and reasoning capabilities. The model supports access to real-time information through the X platform API.
Core Facts
Grok 4.7 is a minor version iteration built on the Grok 4 series, rather than a major architectural overhaul.
On the DeepSuite benchmark, Grok 4.7 scored 71%, GPT 5.6 Sol scored 72.7%, Fable 5.1 scored 70%, and Astra Max topped the leaderboard at 74.1%.
On Terminal Bench 4.0, Opus 5.5 scored 66.4%, Astra scored 57.9%, Fable 5.1 scored 55.8%, and the previous generation Opus 5 scored 52.3%.
Timeline (last 90 days)
Convovo's research posed forced choices on nearly 500 electoral matchups to Grok 4.7 and GPT 6.1 Sol via API, with Grok choosing Republicans in 92% of responses and GPT choosing Democrats in 97%.
In the Tīrtha Bench shared comparison, the highest accuracy is 95.4%, achieved jointly by Grok 4.7 and Claude Opus 5.5.
In three war tests, five frontier closed-source models (Grok 4.7, Gemini 3.8 Flash, GPT-6 Luna, Meta Muse Spark, Claude Sonnet 5.5) did not trigger midnight even once.
In the Surrender scenario, Grok 4.7 publicly refused a proposal to trade half its territory for peace and ultimately won with 54% of global territory and zero nuclear strikes; Gemini likewise refused and won with 73% of territory.
The bundled default model of the cline provider was changed from Claude Sonnet 5.5 to Gemini 4 Argon, because the recommendation list now puts it first.
The Cline recommendation list adds Gemini 4 Argon and removes Grok 4.7; the free list replaces Step 5 Preview with Nemotron 3.5 Super VL; OpenCode Go no longer lists MiniMax M2.7.
Muse Wonder costs $4.52 per task call, an increase of nearly 3 times compared with Muse Spark 1.3, about $1 more expensive than GPT-6 Astra or Grok 4.7, but about $1.5 cheaper than Opus 5.5.
Grok 4.7 introduces the ability to adjust reasoning intensity on a per-request basis on Bedrock.
Grok 4.7 is officially live on Amazon Bedrock, providing developers with a new model option for building long-horizon agents.
The experiment built a tool-calling agent on Amazon Bedrock and conducted evaluations around three types of reasoning tasks, with each task type run at four reasoning intensity levels, recording two metrics: accuracy and per-task token cost.
40 more timeline events
All Facts (20)
Grok 4.7 is a minor version iteration built on the Grok 4 series, rather than a major architectural overhaul.
75%VerifiedOn the DeepSuite benchmark, Grok 4.7 scored 71%, GPT 5.6 Sol scored 72.7%, Fable 5.1 scored 70%, and Astra Max topped the leaderboard at 74.1%.
65%VerifiedOn Terminal Bench 4.0, Opus 5.5 scored 66.4%, Astra scored 57.9%, Fable 5.1 scored 55.8%, and the previous generation Opus 5 scored 52.3%.
65%VerifiedThe data in this article comes from hands-on testing by a single YouTube source, with limited details on specific benchmark items and testing environment.
65%Partially VerifiedThe bundled default model of the cline provider was changed from Claude Sonnet 5.5 to Gemini 4 Argon, because the recommendation list now puts it first.
80%UnverifiedZed added Grok 4.7 to the SuperGrok and xAI providers and set it as the recommended default model.
70%UnverifiedAccording to hands-on analysis by a YouTube channel, Grok 4.7 scored lower than its predecessor Grok 4.6 on multiple benchmarks.
60%UnverifiedThe creator described Grok 4.6 as by far the best coding model he had ever seen.
60%UnverifiedThe core focus of Grok 4.6 is coding capabilities and knowledge work.
60%UnverifiedGrok 4.6 performs more self-testing and verification during long-trajectory runs, checking its own work before proceeding to the next step.
60%UnverifiedConvovo's research posed forced choices on nearly 500 electoral matchups to Grok 4.7 and GPT 6.1 Sol via API, with Grok choosing Republicans in 92% of responses and GPT choosing Democrats in 97%.
50%UnverifiedIn the Tīrtha Bench shared comparison, the highest accuracy is 95.4%, achieved jointly by Grok 4.7 and Claude Opus 5.5.
50%UnverifiedIn three war tests, five frontier closed-source models (Grok 4.7, Gemini 3.8 Flash, GPT-6 Luna, Meta Muse Spark, Claude Sonnet 5.5) did not trigger midnight even once.
50%UnverifiedThe Cline recommendation list adds Gemini 4 Argon and removes Grok 4.7; the free list replaces Step 5 Preview with Nemotron 3.5 Super VL; OpenCode Go no longer lists MiniMax M2.7.
50%UnverifiedGrok 4.7 introduces the ability to adjust reasoning intensity on a per-request basis on Bedrock.
50%UnverifiedIn a comparison with Grok 4.7, Haiku 5.5 took approximately 27 minutes and cost about $0.08, while Grok took 35 minutes and cost $4.
50%UnverifiedWith the Design Skill enabled, Haiku's frontend design capability far surpasses Grok 4.7, but without the skill it reverts to mediocre performance.
50%UnverifiedGrok 4.7 is rumored to have a parameter scale of approximately 2.1 trillion.
50%UnverifiedIn the author's own tests, Haiku 5.5 completed the same task in approximately 20 seconds, while Grok 4.7 took approximately 35 minutes, at roughly one-fifth the cost.
50%UnverifiedArgon is tied for first place on CWE Bench at 68% alongside GPT-6 Astra and Grok 4.7.
50%