agi.score leaderboard
Robust all-source benchmark score
Fresh benchmark evidence from Artificial Analysis, LiveBench 2.0, SWE-bench and Epoch AI. Compare exact model settings, task dimensions and source coverage.
Historical agi.score tools preserved below
Current benchmark results
Compare published results on their own scales. Choose the benchmark and dimension that matches your work; expand a model for its exact source identifier, settings and coverage.
How to read this update
Each source has its own ranking. LiveBench 2.0 is a new release with seven suites; its scores are not mixed with the old January release. Artificial Analysis uses Intelligence Index v4.3.2; publisher-estimated scores are marked and hidden by default. Overall and component scores are alternate views, never duplicate contributions to another average.
Versions and settings stay separate. We preserve publisher identifiers, reasoning effort, fallback variants and harness versions. Similar family names do not prove equivalent configurations. Missing results stay unavailable; published zeroes remain zero. SWE-bench is supplementary evidence, evaluated 20 July 2025 to 26 February 2026; its latest available harness version is selected by default because v1 and v2 are not necessarily comparable.
No new cross-source headline is calculated. Repeated datasets, incompatible releases, uncertain model aliases and Epoch's best scores across settings do not support a defensible blend. Epoch mixes independently run and developer-reported evidence; its 90% intervals are shown. Artificial Analysis marks SciCode under review, so its published composite can change after corrected reruns. Single-run LiveBench and SWE-bench snapshots do not provide equivalent statistical intervals.
The previous robust normalization, source balancing, heuristic evidence shrinkage and Legacy view are preserved below for continuity. The old shrinkage factor is not a calibrated statistical confidence interval. Check current settings, source coverage and the relevant task dimension before making a purchase.
Other current publisher views
Arena text and Arena WebDev coding were dated 22 October 2026; Arena Agent was dated 23 October. These measure different kinds of preference or interaction outcomes. Follow the publisher for intervals, votes and settings. Automated access returned HTTP 403, so no Arena scores were imported or inferred.
Historical combined ranking and source tables
What is scored: LiveBench Global Average plus seven category averages, every numeric MiMo image cell, every numeric Cursor image cell, every numeric cell from the pooled vendor launch comparison tables (Google Gemini 3.5 and Anthropic Opus 4.8), the Artificial Analysis Intelligence Index, and the Liquid AI, StepFun, NVIDIA Nemotron 3 Ultra, Anthropic Mythos 5 / Fable 5, Cohere North Mini Code, GLM-5.2, VibeThinker 3B, and Rio 3.5 Open benchmark cells. The headline score is now robust/source-balanced rather than raw cell-weighted. Rank metrics such as FrontierSWE are inverted so lower ranks score higher.
Missing dashes are excluded from that source/model average rather than treated as zero. Image-derived models are matched to the closest LiveBench variant where the label clearly refers to the same model family.
LiveBench context: objective ground-truth answers, no LLM judge, regularly refreshed questions to limit contamination.
Official source: livebench.ai
Combined Ranking
| # | Model | Organization | agi.score v2 | Legacy | Sources | Cells |
|---|
Methodology
Headline score is agi.score v2. Each benchmark is first converted onto a robust comparable scale instead of forcing every small source table through min-max. Percent metrics use a continuity-corrected logit transform, lower-is-better ranks are inverted, and off-scale point/Elo metrics keep their natural order before robust normalization.
Robust per-benchmark normalization. For each benchmark column, values are centered by the median and scaled by MAD/IQR rather than by the best and worst model in that source. Extreme z-scores are clamped to +/-3 and mapped back to a readable 0-100 score with a sigmoid. This avoids the old failure mode where a tiny four-model launch table automatically gave one model 100 and another 0.
Source-balanced model score. Each source contributes one source score per model, then model scores are averaged across sources rather than cell-weighted across every raw benchmark cell. Evidence confidence is still shown in the final number: confidence = n_eff / (n_eff + 8), where n_eff caps each source at 12 cells, and low-evidence models are shrunk toward 50. The Legacy column preserves the previous raw/min-max cell-weighted score for continuity.
Variant matching and source coverage. Image-derived labels are matched to the closest LiveBench/source equivalent only where the label clearly refers to the same model family or effort setting. Missing dashes remain excluded. Current sources include LiveBench, MiMo, Cursor, vendor launch tables, Artificial Analysis, Liquid AI, StepFun, Nemotron, Mythos, North Mini Code, GLM-5.2, VibeThinker 3B, VibeThinker LeetCode contests, and Rio 3.5 Open.
Source 1 - LiveBench 2026-01-08
| # | Model | Organization | Global | Reasoning | Coding | Agentic | Math | Data | Language | IF | Robust |
|---|
Source 2 - MiMo-V2.5-Pro image comparison
MiMo extracted numeric cells
Source 3 - Cursor Composer 2.5 image comparison
Cursor extracted numeric cells
Source 4 - Vendor launch benchmark tables
Pooled launch-table numeric cells
Source 5 - Artificial Analysis (live, retrieved 25 Oct 2026)
AA effort setting used per model
Artificial Analysis extracted numeric cells
Source 6 - Liquid AI LFM2.5-8B-A1B benchmarks
LFM2.5 extracted numeric cells
Source 7 - StepFun Step 3.7 Flash benchmarks
Step 3.7 Flash extracted numeric cells
Source 8 - NVIDIA Nemotron 3 Ultra keynote comparison
Nemotron 3 Ultra extracted numeric cells
Source 9 - Anthropic Claude Mythos 5 / Fable 5 launch comparison
Mythos 5 extracted numeric cells
Source 10 - Cohere North Mini Code image comparison
North Mini Code extracted numeric cells
Source 11 - Z.ai GLM-5.2 image comparison
Z.ai GLM-5.2 extracted numeric cells
Source 12 - VibeThinker 3B reasoning image comparison
IMO-AnswerBench frontier band
Small and large reasoning tables
Open-source and proprietary comparison
VibeThinker reasoning extracted numeric cells
Source 13 - VibeThinker 3B LeetCode contest image comparison
LeetCode contests (Python)
VibeThinker LeetCode extracted numeric cells
Source 14 - Rio 3.5 Open 397B image comparison
Rio 3.5 Open 397B extracted numeric cells
Source 15 - Anthropic Claude Sonnet 5 launch comparison
Claude Sonnet 5 extracted numeric cells
Citation
@inproceedings{livebench,
title={LiveBench: A Challenging, Contamination-Free {LLM} Benchmark},
author={Colin White and Samuel Dooley and Manley Roberts and Arka Pal and Benjamin Feuer and Siddhartha Jain and Ravid Shwartz-Ziv and Neel Jain and Khalid Saifullah and Sreemanti Dey and Shubh-Agrawal and Sandeep Singh Sandha and Siddartha Venkat Naidu and Chinmay Hegde and Yann LeCun and Tom Goldstein and Willie Neiswanger and Micah Goldblum},
booktitle={The Thirteenth International Conference on Learning Representations},
year={2025},
}Sources & method (refresh 25 Oct 2026)
New in this refresh: three live sources were added and the Artificial Analysis source was replaced with current data. Every number below was downloaded from the linked source on 25 Oct 2026. Nothing was estimated or filled in.
- Artificial Analysis (artificialanalysis.ai), retrieved 25 Oct 2026 04:28 GMT through the AA data API. It now has 49 models and up to 11 columns: the Intelligence, Coding and Math indexes plus the published evals (GPQA Diamond, HLE, SciCode, IFBench, AA-LCR, Terminal-Bench Hard and v4.0, tau2). Fractions are shown as percentages. AA lists several effort settings per model, so the highest-Intelligence-Index setting is used and named in the table.
- LiveBench 2.0 (livebench.ai), release 2026-10-14, the latest public 2.0 release, retrieved 25 Oct 2026. It covers 53 models across seven category scores: Long Agentic, Reasoning, Coding, Browser, Data Analysis, Research and Design. Source 1 below is still the LiveBench 1.0 (2026-01-08) table that the scoring is built on.
- Epoch AI Capabilities Index (epoch.ai/benchmarks, CC-BY), data export retrieved 25 Oct 2026. It covers the 120 models released on or after 2025-06-01, with the latest model release on 2026-10-15. ECI is an off-scale index, so it is min-max normalised like the Elo columns.
Method is unchanged: agi.score v2 still uses robust per-benchmark scaling, source-balanced averaging and evidence shrinkage. Each new source counts as one more source. Names are matched across sources only when they clearly refer to the same model; other models rank on their own evidence.