General
Coding
Agents
Math
Vision
Search
Writing
Language: English
Live data for 2026-10 (recalculated daily)
· Methodology v1
· Methodology
· History
· JSON
#
Tier
Model
fedi-index
Percentile
Value
Sources
Score decomposition
14 A GLM 5.3 Flashprovisional Zhipu AI · Open weights 1,460 ±6 88 +105◆ A ›
25 B Kimi K2.6provisional Moonshot AI · Open weights 1,444 ±4 79 +44 A ›
31 B Gemma 4 31Bprovisional Google · Open weights 1,438 ±4 73 +59 A ›
35 B Qwen3.8 27Bprovisional Alibaba · Open weights 1,433 ±6 70 +46 A ›
36 B Kimi K2.5provisional Moonshot AI · Open weights 1,432 ±4 69 +36 A ›
41 B Qwen3.5 397B A17Bprovisional Alibaba · Open weights 1,428 ±3 64 +27 A ›
43 B MiMo-V2.6-Proprovisional Xiaomi · Open weights 1,425 ±9 62 +93◆ A ›
46 B Gemma 4 26B A4B provisional Google · Open weights 1,423 ±4 60 +54 A ›
47 B MiMo-V2.6-Flashprovisional Xiaomi · Open weights 1,423 ±9 59 +98◆ A ›
49 B MiniMax M3provisional MiniMax · Open weights 1,420 ±4 57 +50 A ›
52 B MiMo-V2.5provisional Xiaomi · Open weights 1,415 ±4 54 +75 A ›
54 B Qwen3 VL 235B A22B Instructprovisional Alibaba · Open weights 1,412 ±4 53 +27 A ›
55 B Qwen3.5-122B-A10Bprovisional Alibaba · Open weights 1,410 ±4 52 +14 A ›
57 B Qwen3.5-27Bprovisional Alibaba · Open weights 1,406 ±3 50 +14 A ›
60 C Inkling Smallprovisional Thinking Machines Lab · Open weights 1,402 ±5 47 +38 A ›
64 C Mistral Medium 3.5provisional Mistral AI · Open weights 1,394 ±6 44 -19 A ›
69 C Qwen3 VL 235B A22B Thinkingprovisional Alibaba · Open weights 1,376 ±8 39 -17 A ›
84 C GLM 4.6Vprovisional Zhipu AI · Open weights 1,341 ±9 26 -12 A ›
88 D GLM 4.5Vprovisional Zhipu AI · Open weights 1,329 ±8 22 -60 A ›
90 D Mistral Small 3.2provisional Mistral AI · Open weights 1,317 ±6 20 -14 A ›
…
A Arena (human votes) ◆ Pareto frontier: nothing is both better and cheaper
fedi-index is an Elo-equivalent on a fixed scale: comparable between months within one methodology version. Tiers S–D: by percentile in the slice, and no better than the confidence interval allows (cut-offs in the methodology). This slice has one source family, so all its scores are provisional.
Methodology
Embed on your site
The table as a widget for your site: it updates by itself and keeps the attribution of the sources.
Language
English
Español
Français
Português (Brasil)
Deutsch
Italiano
繁體中文
Русский
日本語
Bahasa Indonesia
한국어
العربية
ไทย
Türkçe
Polski
Deutsch (Schweiz)
Norsk
Slovenščina
Dansk
Svenska
Nederlands
Ελληνικά
Suomi
Română
Theme
Automatic
Light
Dark
Height, px
Code
Copy code
Copied
JSON
fedi-index © fedi.software, CC BY 4.0; components keep the licences of their sources.
fedi-index © fedi.software, CC BY 4.0; components keep the licences of their sources. · Sources:
LMArena (CC BY 4.0),
Epoch AI (CC BY 4.0),
τ²-bench (Sierra) (MIT),
models.dev (MIT),
LiteLLM (MIT)
Arena Leaderboard Dataset by LMArena, CC BY 4.0. Modified by fedi.software (equated, weighted, aggregated). Not endorsed by LMArena.
Epoch AI, 'Capabilities & benchmarking'. Published online at epoch.ai. Retrieved from 'https://epoch.ai/benchmarks'. CC BY. ECI by Epoch AI; Epoch-run results only. Modified by fedi.software.
τ²-bench leaderboard, © Sierra Research, MIT License (Yao et al. 2024, arXiv:2406.12045; Barres et al. 2025, arXiv:2506.07982). Sierra-run text submissions only.
Pricing: models.dev, MIT License, © 2025 models.dev.
Pricing: LiteLLM model_prices_and_context_window.json, MIT License, © 2023 Berri AI.