Independent AI research
Vol. 01 — Nov 15, 2026Every task has a different winner.
Noesira ranks AI models by what they're actually best at — writing, coding, reasoning, research, routing — compiled from public benchmarks and arenas, published with the sources attached.
Highest measured score on SWE-bench Verified — Claude Opus 5
Task categories ranked in our first published study
Different labs hold the #1 spot across the ten headline categories
Median decision time of Jev 1.14, the fastest top-ranked routing model
The board — Nov 15, 2026
Who's winning what
Every row links to its full table with sources. Where boards disagree, we publish both — start with the study.
There is no best model. There is a best model for this question — and the honest way to find it is to measure everything, publish the numbers, and show the sources.
Overall
The top 10, all-in
Artificial Analysis Intelligence Index v5 — the closest thing the field has to a general ranking. Useful as a map; misleading as a verdict. The per-task tables are where decisions get made.
Intelligence Index — top 10
Artificial Analysis · Nov 15, 2026| # | Model | Index | Output price |
|---|---|---|---|
| 1 | Claude Fable 5.2 (Max)Anthropic | 65 | $26.54 /M |
| 2 | Claude Fable 5.2 (Xhigh)Anthropic | 62 | — |
| 3 | Gemini 4 Carbon (High)Google | 60 | $5.48 /M |
| 4 | Claude Opus 5.5 (Max)Anthropic | 59 | — |
| 5 | GPT-6 Astra (Max)OpenAI | 57 | — |
| 6 | GPT-6.1 Sol (Max)OpenAI | 55 | $1.27 /task |
| 7 | Claude Sonnet 5.5 (Max)Anthropic | 55 | — |
| 8 | Gemini 4 Hydrogen (Xhigh)Google | 54 | — |
| 9 | GPT-6.1 Terra (Max)OpenAI | 51 | $0.96 /task |
| 10 | Gemini 3.1 Pro PreviewGoogle | 48 | $2.00 /M in |
Full tables, every category: Rankings →
Compiled from the public record
The routing finding
Dedicated routing models are here — and the independent numbers are humbling
Google's Gemini Decision 1 launched November 11 with best-in-suite claims. On Artificial Analysis's independent Decision Index it scores 66 — exactly the field median — while TypeSafe's Jev 1.14 edges it on score, runs 2.3× faster, and costs less than half. Vendor claims and independent measurement are telling two different stories. We track both.
Decision Index
Median latency
Cost / 1M decisions
Research
Latest publication
001Which model is best at what
The per-task leaders across twelve categories, from 15 sourced research queries — including where the leaderboards flatly disagree with each other, and what that means for anyone routing between models.
About
An independent research company, not a leaderboard farm
Noesira is the sister company to Discera, the chat that routes every question to the model best suited to it. Our rankings feed Discera's routing — and because they're published here, with sources, anyone can check the work. No lab funds this research and no placement is for sale. Evaluation services for other teams come later; the public research program comes first.