MODEL LAUNCHES / FOLLOW THE EVALUATION

What the latest models are measured on.

A model launch is a collection of claims under specific conditions. Explore the benchmarks behind the announcement, then read the protocol before comparing the scores.

Kimi

Kimi K3.1 Pro

34 linked benchmarks

Explore the evaluation suite →MiniMax

MiniMax M3.1 Flash

6 linked benchmarks

Explore the evaluation suite →DeepSeek

DeepSeek V4.1 Flash 1017

11 linked benchmarks

Explore the evaluation suite →Anthropic

Claude Opus 5.5

9 linked benchmarks

Explore the evaluation suite →OpenAI

GPT-6 Astra

25 linked benchmarks

Explore the evaluation suite →Meta

Muse Wonder and Muse Wonder Full

15 linked benchmarks

Explore the evaluation suite →Google

Gemini 4 Argon

18 linked benchmarks

Explore the evaluation suite →Kimi

Kimi K2.6 · earlier-2026 baseline

25 linked benchmarks

Explore the evaluation suite →Qwen

Qwen3.5 · earlier-2026 baseline

26 linked benchmarks

Explore the evaluation suite →

Launch reports are primary evidence for what a lab reports. Inclusion does not turn a lab-run result into an independent evaluation. Reviewed 26 October 2026.