Grok 4.5
A large language model in the Grok series released by xAI, included in comparative benchmarking evaluations of AI coding capabilities.
Timeline (last 90 days)
Grok 4.5's performance on Tīrtha Bench was 93.4% accuracy, a 98.3% trap pass rate, 1.85 seconds latency, and $0.0015 per correct answer.
The September 18 changelog set October 19 as the removal date for another six models, including GPT-5.5, GPT-5.4, and Grok 4.5; the interval from October 19 to November 16 is only 28 days.
Questions for which all retries still failed were scored as incorrect rather than discarded; the author speculates that Grok 4.5's score of 0.00 was most likely caused by an infrastructure failure, but this has not been conclusively confirmed.
Musk hyped the release of Grok 4.7 for more than a month and claimed at the time of Grok 4.5's launch that 4.7 would surpass all current models.
Model selection strategy: use Composer 2.5 for simple web tasks, Grok 4.5 for moderately challenging tasks, and switch to advanced models such as GPT 5.6 or Fable when the first two cannot handle the task.
The Grok 4.5 model was trained in collaboration with SpaceX using its computing resources directly, and its price per token is much cheaper than comparable models such as GPT 5.6.
Grok 4.6 performs excellently across various benchmarks, with a per-token price equal to that of 4.5.
Zed v1.18.0 adds Gemini 3.5 Flash-Lite on the Google AI side, and Grok 4.5 and Grok 4.6 on the xAI side.
Grok 4.6 is an iterative upgrade (dot upgrade) based on Grok 4.5, not a newly trained model.
The user stated that Grok 4.6 consumes quota much faster than Grok 4.5.
6 more timeline events
All Facts (7)
Grok 4.5's performance on Tīrtha Bench was 93.4% accuracy, a 98.3% trap pass rate, 1.85 seconds latency, and $0.0015 per correct answer.
50%UnverifiedQuestions for which all retries still failed were scored as incorrect rather than discarded; the author speculates that Grok 4.5's score of 0.00 was most likely caused by an infrastructure failure, but this has not been conclusively confirmed.
50%UnverifiedModel selection strategy: use Composer 2.5 for simple web tasks, Grok 4.5 for moderately challenging tasks, and switch to advanced models such as GPT 5.6 or Fable when the first two cannot handle the task.
50%UnverifiedThe Grok 4.5 model was trained in collaboration with SpaceX using its computing resources directly, and its price per token is much cheaper than comparable models such as GPT 5.6.
50%UnverifiedGrok 4.6 is an iterative upgrade (dot upgrade) based on Grok 4.5, not a newly trained model.
50%UnverifiedThe September 18 changelog set October 19 as the removal date for another six models, including GPT-5.5, GPT-5.4, and Grok 4.5; the interval from October 19 to November 16 is only 28 days.
50%UnverifiedZed v1.18.0 adds Gemini 3.5 Flash-Lite on the Google AI side, and Grok 4.5 and Grok 4.6 on the xAI side.
50%