We rechecked the Artificial Analysis (AA) Intelligence Index and the LMArena text board on October 27. Claude Opus 5.5 still leads, at 58 on max effort. Claude Sonnet 5.5 (max) and Meta's Muse Wonder Full (max) follow on 56, then Moonshot's Kimi K3.1 Pro on 55. Claude Fable 5.1, GPT-6 Astra and Gemini 4 Argon sit together on 53.

New and updated rows

Benchmarks now includes Claude Sonnet 5.5 (56), Kimi K3.1 Pro (55), GPT-6.1 Sol (52), GPT-6.1 Terra (48) and Claude Haiku 5.5 (43). Three existing rows moved from a gap or an older reading to AA's own figure: Qwen3.8-Max 45 (AA tests its 0902 snapshot), DeepSeek V4.1-Flash 45 (AA tests the 1017 checkpoint that DeepSeek's API now serves; we had 39) and the Mistral Large 4 preview 38.

AA lists several effort levels per model. We show each model's best listed variant and label it, so a 'max' row and a 'high' row are not like-for-like on cost.

LMArena

On the Arena text board's October 26 refresh, as relayed by AI Blogly, Gemini 4 Argon (High) leads at 1517, ahead of Claude Opus 5.5 (High) and Muse Wonder (Max) at 1509 and Claude Fable 5.1 (Max) at 1503, which we had listed as a gap. Sonnet 5.5 is not in the top ten of the overall text board.

Corrections made during this refresh

Rechecking sources, we found list prices and context windows on this site that did not match the labs' own pages. Claude Opus 5.5 is $4/$20 and Sonnet 5.5 is $2/$10 on Anthropic's pricing page; we had $5/$25 and $3/$15. GPT-6.1 Sol is $2/$10 on OpenAI's; we had $5/$30. Fable 5.1 has a 1M-token context, not 500K, and the GPT-6 family lists 1.05M. DeepSeek's current Flash model lists 1M and Qwen3.8-Max about 1M. Mistral Large 4 states no context window: Artificial Analysis lists 524K and OpenRouter 1M. Affected stories carry a correction note.