Weighted consensus · Public benchmarks

LLM leaderboard

Name: UnifyBench public ranking snapshot
Creator: UnifyBench
Published: 2026-10-20T00:24:26.243Z
License: https://unifybench.ai/data#reuse

Compare AI models across public benchmarks. Source-balanced consensus for overall performance, coding, reasoning and agentic work.

Looking for which LLM is best? UnifyBench is a first stop: source-balanced consensus across many published benchmarks (not one board). Shortlist here, then validate on Artificial Analysis, check Arena for preference, and finish with task-specific evals.

Overall leaders

Comparison score · 0–100

Consensus across available benchmarks. Models cover different task areas; use a task ranking for a focused comparison.

Experimental scores for the selected profile and sources. Small gaps may not be meaningful. How scoring works · Effort evidence collected 2026-10-18

This month: three boards to compare

Which LLM is best in October 2026?

The dated checks below show different measures of performance: GPT-6 Astra · XHigh leads UnifyBench; Claude Opus 5.5 (Max, Default Fallback) is Artificial Analysis's last known leader; Gemini 4 Argon · High is Arena's last known leader. Shortlist with UnifyBench, check Artificial Analysis for independent scores and cost per task, check Arena for preferences, then test your own tasks.

UnifyBenchGPT-6 Astra · XHigh95.9Overall · benchmark consensus · 0–100
Artificial AnalysisClaude Opus 5.5 (Max, Default Fallback)58Intelligence Index · independent suiteLast known leader
ArenaGemini 4 Argon · High1525 ± 9Text Overall Elo · human preferenceLast known leaderPreliminary

The Intelligence Index, Arena Elo, and UnifyBench score use different scales and cannot be compared directly. UnifyBench balances many published benchmarks rather than blending a few board ranks. Why leaderboards disagree.

Source dates and refresh details

UnifyBench
Evidence published 20 Oct 2026
Default weights · documented evidence · highest-scoring effort

Artificial Analysis
Checked 7 Oct 2026 · Source update date unavailable
#1 · Stale check · Refresh failed 20 Oct 2026; last good check retained

Arena
Checked 7 Oct 2026 · Board updated 2 Oct 2026
#1 · Preliminary · Stale check · Refresh failed 20 Oct 2026; last good check retained

471unique models 568model configurations 134source pages

Explore all 134 source pages Independent evaluations and lab reports

Each model is counted once across effort levels. Configurations count documented model + effort settings, including unranked.

Original reports and benchmark leaderboards. Distinct source pages with published results in the catalog. Multiple results and page sections count as one source.

One entry per model, using its highest-scoring eligible effort setting. How effort ranking works.

Customize ranking Sources, profiles and evidence details

Full ranking

231 entries

Compare two configurations

Capability ranking of measured model configurations. Scores are experimental; missing results remain unknown.
Rank	Model	Score	Evidence breadth Why this score?
176	ERNIE-4.5-300B-A47B-PT · Non-reasoningBaidu	0.8	Single independent source11 families · 4 areas measured
177	Qwen3-14B · ReasoningQwen	0.8	Single independent source16 families · 5 areas measured
178	Qwen3-VL-8B-Thinking · ReasoningQwen	0.8	Single independent source12 families · 6 areas measured
179	Gemma 4 E4B IT · Non-reasoningGoogle	0.8	Single independent source9 families · 6 areas measured
180	NVIDIA-Nemotron-3-Nano-4B-BF16 · ReasoningNVIDIA	0.7	Single independent source12 families · 5 areas measured
181	Devstral Small 2 · Non-reasoningMistral	0.6	Single independent source17 families · 6 areas measured
182	NVIDIA-Nemotron-Nano-9B-v2 · ReasoningNVIDIA	0.6	Single independent source11 families · 5 areas measured
183	Mistral Medium 3 · Non-reasoningMistral	0.6	Single independent source13 families · 6 areas measured
184	Amazon Nova Premier · Non-reasoningAmazon	0.6	Single independent source13 families · 5 areas measured
185	Qwen3-Coder-30B-A3B-Instruct · Non-reasoningQwen	0.6	Single independent source12 families · 5 areas measured
186	DeepSeek-R1-Distill-Llama-70B · ReasoningDeepSeek	0.6	Single independent source12 families · 5 areas measured
187	Ministral 3 14B · Non-reasoningMistral	0.5	Single independent source17 families · 6 areas measured
188	DeepSeek-R1-0528-Qwen3-8B · ReasoningDeepSeek	0.5	Single independent source12 families · 5 areas measured
189	DeepSeek-R1-Distill-Qwen-32B · ReasoningDeepSeek	0.5	Single independent source8 families · 4 areas measured
190	DeepSeek-R1-Distill-Qwen-14B · ReasoningDeepSeek	0.5	Single independent source8 families · 4 areas measured
191	Qwen3-Omni-30B-A3B-Instruct · Non-reasoningQwen	0.5	Single independent source12 families · 6 areas measured
More models on this page 192 Qwen3-8B · Reasoning 0.5 193 Qwen3.5-2B · Non-reasoning 0.4 194 Qwen3-4B-Instruct-2507 · Non-reasoning 0.4 195 Mistral Small 3.2 · Non-reasoning 0.4 196 Llama-4-Scout-17B-16E-Instruct · Non-reasoning 0.4 197 Qwen3-VL-8B-Instruct · Non-reasoning 0.4 198 Ministral 3 8B · Non-reasoning 0.4 199 Qwen3-VL-4B-Thinking · Reasoning 0.4 200 Gemma 4 E2B IT · Reasoning 0.4

Behind the ranking

294 entries with published results · 231 ranked · 7 provisional estimates.

Explore seven capabilities, the original results, and gaps in the evidence.

Coding ranking Reasoning ranking Agentic ranking Knowledge ranking Compare capabilities ↗Published benchmarks ↗Evidence coverage ↗

Cite this ranking

Include the model and effort setting, the selected profile, and the dated data downloads. A shared link uses the latest data; it does not freeze a ranking in time.

UnifyBench ranking
Data updated: 20 Oct 2026
Mode: overall; evidence: documented; configurations: best
https://unifybench.ai/?evidence=documented&mode=overall&page=8

Catalog data · Effort measurements · Configuration scoring rules · Methodology

Catalog dated 20 Oct 2026. Effort evidence collected 18 Oct 2026. Collection dates are not evaluation dates.

What does the score mean?

The adjusted comparison score estimates performance against the same reference panel from matched published comparisons. Capability pools matched comparisons across families and checks aggregate breadth and independent source coverage. Custom profiles require support in every selected capability. Missing results never count as losses.

Family details show observed win shares against available reference peers. Those descriptive values are supporting evidence, not adjusted capability scores. Provider reports and independent results remain labeled. Matching reported settings do not establish equal inference effort, and score differences do not establish statistical significance.

Read the full methodology →