New in models & benchmarks

Google released Gemini 3.7 Flash on 2026-08-13 and xAI released Grok 4.6 on 2026-08-12. One now leads GPQA Diamond outright; the other sits top five on Terminal-Bench 2.1 and the AA Coding and Intelligence indices, at very different speeds and prices.

LiveBench rises 8.7 points. Artificial Analysis's stable DeepSeek V4 Flash entry gains 9.7 points on its Intelligence Index, and Epoch AI's WeirdML results rise by 13 to 17 points at matched reasoning settings.

Artificial Analysis's Intelligence Index puts Gemini 3.6 Flash level with the model it replaces, 50.1 against 50.2. Underneath that, agentic coding is up 15.1 points on GBENCH and 11.2 on DeepSWE, social reasoning is down 19.1, and the model runs 35.5% faster for 16.7% less per output token.

Each benchmark row in Customize board now reads coverage against the models you are actually looking at, and Ollama Cloud, Synthetic, and OpenCode Go arrive as predefined model sets.

Artificial Analysis's hallucination-penalized knowledge index lands with first results for 272 models. Claude Fable 5 leads at 40.1, 7.2 points ahead of Gemini 3.1 Pro — and the models least likely to hallucinate are different ones entirely.

Artificial Analysis's first-hand SciCode evaluation now covers 310 models; Claude Fable 5 is 1.3 points ahead of second place.

LiveBench profiles Thinking Machines' debut model mid-pack at its highest effort setting, and Grok 4.5 enters DeepSWE effectively even with Claude Sonnet 5.

A day after release, independent boards post their first Kimi K3 results: first place among 64 models on ClawProBench and a close second of 32 on LiveBench Reasoning.

Across 15 first observations, Kimi K3 places in the top three seven times and leads the newly added AutomationBench-AA board.

Sol ranks first across the four views among 20 models with scores, while Claude Fable 5 remains in the top five on each.

Fable leads more of the 19 added coding and agentic measures than any other model, while GPT-5.6 Terra and Claude Opus 4.8 lead different kinds of work.

At the same maximum effort setting, Sol leads overall by 1.61 points and Agentic Coding by 18.69 points.