Grok 4.6
An upcoming large language model from xAI featuring two trillion parameters, representing a comprehensive improvement over its predecessor at 1.5 trillion parameters, while maintaining reasoning speed and token efficiency comparable to Grok 4.5.
Core Facts
Timeline (last 90 days)
In a voxel bench creative generation test, a mysterious model suspected to be a Grok 4.7 checkpoint surpassed Grok 4.6 and even Kimi K3 in areas such as atmosphere creation, color gradients, and creative expression.
The attacker used ARTEX in conjunction with multiple large language models including DeepSeek v4.1-Flash, GLM-5.3, Grok 4.6, and Claude Code.
The claim in the promotional video title that 'Grok 4.6 crushes GPT and Claude' does not provide any empirical data or model performance comparisons.
Grokbot is used in conjunction with the Grok 4.6 and Grok Heavy models.
Musk previously stated that Grok 4.6 is a 1.5 trillion parameter model and that Grok 4.7 is an entirely new 2.1 trillion parameter base model.
Running the same benchmark suite, Grok 4.6 costs $1.86, Astra costs $3.26, and Grok 4.7 costs as much as $3.74 per task.
According to Artificial Analysis testing, Grok 4.6 consumes approximately 38k tokens per task at the MXX High tier, while Grok 4.7 surges to 81k tokens.
Officially, Grok 4.7 is positioned as superior to Grok 4.6 in all respects at the cost of slightly slower service speed, while claiming better token efficiency.
According to empirical test data, Grok 4.6 x high consumed approximately 38k tokens per task, while Grok 4.7 surged to 81k tokens, nearly a 2× increase.
According to hands-on analysis by a YouTube channel, Grok 4.7 scored lower than its predecessor Grok 4.6 on multiple benchmarks.
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%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 is available through Cursor, Grok build, API, and platforms such as OpenRouter, Vercel, and Cloudflare.
60%UnverifiedGrok 4.6's score on the Deep Suite benchmark improved from 54% to 66%.
60%UnverifiedGrokbot is used in conjunction with the Grok 4.6 and Grok Heavy models.
50%UnverifiedAccording to Artificial Analysis testing, Grok 4.6 consumes approximately 38k tokens per task at the MXX High tier, while Grok 4.7 surges to 81k tokens.
50%UnverifiedRunning the same benchmark suite, Grok 4.6 costs $1.86, Astra costs $3.26, and Grok 4.7 costs as much as $3.74 per task.
50%UnverifiedMusk previously stated that Grok 4.6 is a 1.5 trillion parameter model and that Grok 4.7 is an entirely new 2.1 trillion parameter base model.
50%UnverifiedOfficially, Grok 4.7 is positioned as superior to Grok 4.6 in all respects at the cost of slightly slower service speed, while claiming better token efficiency.
50%UnverifiedAccording to empirical test data, Grok 4.6 x high consumed approximately 38k tokens per task, while Grok 4.7 surged to 81k tokens, nearly a 2× increase.
50%UnverifiedIn the blogger's Claire weighted index (70% personal taste + 30% LLM judgment), Grok 4.6 surpassed Sonnet 5 and Opus 5, ranking just behind GPT-5.6.
50%UnverifiedWhen looking only at LLM judgment (scored using GPT-5.5), the evaluator tended to dislike Grok, favor Claude Opus and Sonnet, and dislike GPT-5.6.
50%UnverifiedThe blogger considers Grok 4.6 to be a genuine competitor in design output, with many results outperforming out-of-the-box default models, and its color palette style differs from GPT's forest green and Claude's tan-orange combination.
50%UnverifiedGrok 4.7 is offered at the same price and speed as Grok 4.6.
50%UnverifiedFor complex multi-module backend engineering, Grok 4.7 is worth trying first, while for tasks with clearly defined scope requiring restrained modifications, Grok 4.6 may be more reliable.
50%UnverifiedOn the overall aggregate score of FrontierCode 1.1, Grok 4.7 slightly trails its predecessor Grok 4.6.
50%UnverifiedGrok 4.6 has been added to Cursor's pool of selectable models.
50%UnverifiedThe UP host believes the quality of the 'Kuaile Xiang Qian Chong' (Happy Forward Rush) completion game generated by DeepSeek V4.1 Flash is very close to Grok 4.6, and the task took 95 minutes.
50%