Superbash
SearchUser Hub

AI model benchmarks

Choose a model by practical tier, then inspect every published same-prompt run behind the recommendation.

S

Claude Fable 5.1AnthropicGPT-5.6 SolOpenAIGrok 4.6Xai

A

Kimi K3Moonshot AIDeepSeek V4 ProDeepSeekGPT-5.6 TerraOpenAIGLM 5.3Z.aiGLM 5.3 FlashZ.aiQwen3.8-MaxAlibaba Qwen

B

Gemini 3.7 FlashGoogleClaude Fable 5AnthropicDeepSeek V4 FlashDeepSeekClaude Opus 5AnthropicGPT-5.5OpenAIMuse Spark 1.2Meta

C

GPT-5.6 LunaOpenAIMiniMax M3MiniMaxGemini 3.1 Pro PreviewGoogleMiMo-V2.5-ProXiaomiClaude Sonnet 5AnthropicQwen3.7-PlusAlibaba QwenNVIDIA Nemotron 3 UltraNVIDIA

D

LongCat-2.0MeituanHy3Tencent Hy TeamStep 3.7 FlashStepFun

Additional benchmark runs

Published test sets that are not part of the current editorial tier ranking.

Composer 2.5CursorGrok 4.5XaiInklingThinking Machines Lab
25 models across 5 tiers
223 runs across 19 tested models

Updated

Model hub next step

Compare the models before you commit.

Start with the editorial tier, then open the source-linked cost and intelligence tables for the workload you actually have. Tier placement is routing guidance, not a substitute for a current evaluation.

Last reviewed

Link map: Compare models and costsInspect test briefsLearn how tiers are set

Evidence and tools

Compare the evidence.

Compare independent scores, API costs, and the same prompts built by different models.

Open benchmark and cost explorerBrowse test briefs

Benchmark hub next step

Turn a model choice into a testable brief.

Use a published prompt to compare the same job across models, then keep the output, method, and caveats together when you make a call.

Last reviewed

Link map: Choose a benchmark briefCompare model evidenceReview the testing skill