Weighted consensus · Public benchmarks

LLM reasoning leaderboard

Compare AI reasoning models and measured effort settings across published reasoning benchmarks.

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.

Reasoning leaders

Comparison score · 0–100

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

Reviewed retained observations are in use. October 2026 Real-SWE board has no 95% confidence whisker, so these scores stay at the 2026-10-18 retrieval. Original source retrieval 2026-10-18; source updated 2026-10-18.

Explore all 137 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

Evidence only MultimodalLong contextHuman prefAll capabilitiesThese profiles do not yet have enough comparable evidence for a ranking.

Adjust weights

Full ranking

332 entries

Custom capability ranking of measured model configurations. Scores are experimental; missing results remain unknown.
RankModelScoreCapability coverage
176Qwen3.5-flash · Non-reasoningQwen8.5Eligible · 100%2 families · 1 selected capabilities
177MiniCPM5-2B · ReasoningOpenBMB8.1Eligible · 100%3 families · 1 selected capabilities
178DiffusionGemma 26B A4B · ReasoningGoogle8.1Eligible · 100%3 families · 1 selected capabilities
179Qwen3-VL-32B-Thinking · ReasoningQwen8.1Eligible · 100%4 families · 1 selected capabilities
180Gemma 4 12B IT · ReasoningGoogle8.0Eligible · 100%3 families · 1 selected capabilities
181GPT-5 Nano · HighOpenAI8.0Eligible · 100%5 families · 1 selected capabilities
182Seed OSS 36B Instruct · ReasoningByteDance7.9Eligible · 100%4 families · 1 selected capabilities
183Llama Nemotron Super 49B v1.5 · ReasoningNVIDIA7.9Eligible · 100%5 families · 1 selected capabilities
184Qwen3-32B · ReasoningQwen7.9Eligible · 100%5 families · 1 selected capabilities
185Magistral Small 1.2 · ReasoningMistral7.9Eligible · 100%4 families · 1 selected capabilities
186Nova 2.0 Omni · MediumAmazon7.9Eligible · 100%4 families · 1 selected capabilities
187GLM-4.6V · ReasoningZ.ai7.7Eligible · 100%4 families · 1 selected capabilities
188Mistral Medium 3.5 · ReasoningMistral7.5Eligible · 100%3 families · 1 selected capabilities
189GLM-4.5-Air · ReasoningZ.ai7.4Eligible · 100%5 families · 1 selected capabilities
190Qwen3-VL-30B-A3B-Thinking · ReasoningQwen7.4Eligible · 100%4 families · 1 selected capabilities
191Claude Haiku 4.5 · ReasoningAnthropic7.2Eligible · 100%4 families · 1 selected capabilities

More models on this page

About this ranking

This profile gives hard reasoning all the weight. It compares identified configurations using their own measured results, rather than borrowing results from another effort setting. Rankings require support in this capability. Matching effort labels across providers do not imply equal compute budgets.

View the Capability ranking →

Behind the ranking

416 entries with published results · 332 ranked · 0 provisional estimates.

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

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: 22 Nov 2026
Mode: custom; evidence: documented; configurations: best
Weights: v3;agentic:0,hard_reasoning:100,coding:0,human_pref:0,knowledge:0,multimodal:0,long_context:0
https://unifybench.ai/rankings/reasoning?evidence=documented&page=8&view=shortlist

Catalog data, effort measurements, and scoring rules · Methodology

Catalog dated 22 Nov 2026. Effort evidence collected 22 Nov 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 →