Skip to main content
BenchLM
Data

LLM Leaderboard & AI Model Benchmarks — November 2026

As of November 21, 2026, the BenchLM AI leaderboard ranks 216 models. Claude Fable 5.2 is first by point estimate. Conditional ranges do not establish rank confidence. Open the evidence before choosing.

Search models and benchmarks

Current BenchAlign leaders

Snapshot November 21, 2026

Method
  1. Anthropic logo

    Anthropic01

    Score93.390% interval 88.5–98.1
    Agentic
    93
    Coding
    92
    Reasoning
    94
    Multimodal
    87
    Knowledge
    93
    Multilingual
    88
    Instruction following
    88
    Math
    87

    Supportedenough independent evidence

    Details for Claude Fable 5.2
  2. Google logo
    Gemini 4 NeonSupported

    Google02

    Score91.3Conditional range 85.0–97.5
    Agentic
    92
    Coding
    93
    Reasoning
    89
    Multimodal
    81
    Knowledge
    88
    Multilingual
    Not measured
    Instruction following
    87
    Math
    84

    Supportedenough independent evidence

    Details for Gemini 4 Neon
  3. OpenAI logo
    GPT-6.2 AstraSupported

    OpenAI03

    Score90.8Conditional range 84.5–97.2
    Agentic
    93
    Coding
    92
    Reasoning
    87
    Multimodal
    90
    Knowledge
    89
    Multilingual
    Not measured
    Instruction following
    83
    Math
    84

    Supportedenough independent evidence

    Details for GPT-6.2 Astra

Price context: Gemini 4 Carbon retains 95% of the top score with an output price 75% lower.

Data refreshed
November 21, 2026
Supported
147 models
Ranking method
BenchAlign v6.0

The LLM leaderboard — November 2026 rankings

Score, model size, context, price, and benchmark detail in one table. Filter state syncs to the URL so every view is shareable. Supported and Estimated labels show how much independent evidence backs each position.

Supported means we have enough independent evidence to stand behind the row. Estimated means the rank is visible and the uncertainty is wider.

Supportedenough independent evidenceEstimatedwider uncertainty

Parameters show total billions; active counts appear when a source reports them. Size filters use total parameters: Tiny ≤4B, Small >4–<40B, Medium 40–150B, and Large >150B.

AG Agentic CO Coding RE Reasoning MM Multimodal & Grounded KN Knowledge ML Multilingual IF Instruction Following MA Math

Showing 25 of 1035 models · 216 ranked

Source-linked · the table lists all 1035 tracked models; 216 carry a BenchAlign rank

How the score is built
Describe the task and any limits. You get a ranked shortlist. It does not carry out the task.

We keep what you type, drafts included, for 180 days, with no IP address or account attached. Leave out personal details. Privacy

What the table covers

The BenchLM LLM leaderboard 2026 ranks 216+ models and tracks 1035+ large language models side by side across 835 benchmarks — from SWE-bench and LiveCodeBench for coding to GPQA Diamond and MMLU-Pro for knowledge and reasoning. Whether you need the best AI models 2026 has to offer for agentic workflows, math, multilingual tasks, or instruction following, our AI benchmark comparison tables make it easy to see how GPT-5, Claude, Gemini, DeepSeek, Llama, and dozens of other frontier and open-source models stack up on both benchmarks and operator tradeoffs like price and context. Supported and Estimated labels separate evidence strength from score, so incomplete reporting does not automatically remove a model from the comparison.

What changed in November 2026

  1. Claude Fable 5.2 (Anthropic) holds the #1 spot on the BenchAlign leaderboard at 93.3.
  2. 32 model releases tracked in November 2026.
  3. Google climbed 2 spots in the provider race.
  4. 147 Supported and 69 Estimated models appear in the public leaderboard.

Dataset journalData verified

Build a head-to-head comparison

Choose any two ranked models. The comparison opens with the decision, exact units, and evidence coverage.

Decision-ready picks

Current models that clear the ranking, freshness, and evidence thresholds for each decision.

DecisionModelCurrent measureProvenance
  • Best open weightHighest-scoring evidence-qualified current open-weight model.Kimi K3.1 ProMoonshot AI70overall scoreSupported · 26 source families
  • Best near-frontier value95% of the leading score at a lower output price.Gemini 4 CarbonGoogle$10output / 1M tokensSupported · 39 source families
  • Fastest measuredFastest externally measured model that also clears the ranking and evidence thresholds.LFM2.5-8B-A1BLiquidAI335tokens / secArtificial Analysis · updated 2026-11-20
  • Largest useful contextLargest context among evidence-qualified current models retaining at least half of the leading score.GPT-6.2 AstraOpenAI1.05Mcontext windowSupported · 25 source families

Current models that clear rank, freshness and evidence thresholds

Qualification method

How we work

A ranking is a shortlist.

Rank orders the evidence; it does not make the decision. Every row here opens to its sources so you can disagree with us.

We do not invent scores to fill a table.

Verified launch results can inform an Estimated rank, labeled as such, while Supported ranks require independent evidence. Models without verifiable public results stay unranked.

When we were wrong, the fix is on the page.

Scoring bugs, missing coverage, and re-weightings are dated and linked, because a ranking you cannot audit is a launch post with a table.

The AI race, read as a record instead of a spectacle

Track which release led its month, how provider depth changed, and whether the crown moved.

Months
6
Releases
317
Lead changes
5

Best ranked model released in Nov 2026

Claude Fable 5.2Anthropic
Score93.3
RankProviderTop-three avg.Models
01Google86.222
02OpenAI85.639
03Anthropic84.924

Six-month release record

Explore the release timeline

Scoring methodology

Scoring methodology6 core categories in Overall, 2 separate category boards, 9 specialist verticals, 113 scored benchmarks, and 356 display-only records.Inspect method
Overall weights
100%
Overall categories
6
Scored benchmarks
113
Display-only
356

Overall combines six core categories: Agentic (25%), Coding (25%), Reasoning (20%), Knowledge (15%), Math (10%), and Instruction Following (5%), plus open CC BY 4.0 preference ratings capped at 10%. Within each category, tests of unequal difficulty are calibrated onto one scale so missing tests widen uncertainty instead of counting as zero.

Multimodal & Grounded and Multilingual have separate calibrated rankings and carry 0% weight in Overall, so text-only and English-focused models are not penalized. Nine specialist vertical leaderboards also stay separate.

Scores come from benchmark-native public leaderboards, independent evaluators, and verified provider launch cards. Verified launch results can inform an Estimated rank at reduced weight (0.72, capped at 2.2 per category), while Supported ranks require independent evidence across multiple benchmark families.

Read the complete methodology
Category and Overall shareBenchmarks that affect the scoreCoverage
  • Agentic25%

    DRACO · Terminal-Bench 4.1 · Terminal-Bench 4.0 · Terminal-Bench-Science 0.1 · Terminal-Bench 2.1 · BrowseComp · HLE w/ tools · Gert Labs · OSWorld-Verified · OSWorld 2.1 · OSWorld 2.1 Offline · JobBench · AndroidWorld · MCP Atlas · AutomationBench · Agents' Last Exam · τ²-bench results · DeepSearchQA · PinchBench · SWE Refactor Bench · BFCL v4 · Claw-Eval · ResearchClawBench · QwenClawBench · τ³-bench results · VITA-Bench · DeepPlanning · WideResearch

    28 scored · 117 context-only

  • Coding25%

    PostTrainBench v1.1 · PostTrainBench v1 · SWE-Rebench · LiveCodeBench v6 · LiveCodeBench Pro · SWE-bench Pro · VulcanBench Frontier v4 · SWE Multilingual · FrontierSWE · FrontierSWE v2 · Bug Hunt Bench · PostTrainBench · NL2Repo · React Native Evals · SciCode · EEBench · DeepSWE

    17 scored · 57 context-only

  • Reasoning20%

    STEMBench · GraphWalks BFS 256K–1M · LongBench v2 · MRCRv2 · ARC-AGI-2

    5 scored · 33 context-only

  • Knowledge15%

    BioMysteryBench (human-difficult) · GPQA-D · SuperGPQA · MMLU-Pro · HLE · HLE-Verified · LABBench2 · HealthBench Hard · HealthBench Professional · MedXpertQA (Text) · HLE w/o tools · MMLU-Redux · C-Eval

    13 scored · 54 context-only

  • Math10%

    AIME 2025 · AIME26 · HMMT Feb 2025 · HMMT Nov 2025 · HMMT Feb 2026 · IMOAnswerBench · Apex · MMAnswerBench · FrontierMath v2 (Tiers 1-3) · FrontierMath v2 (Tier 4)

    10 scored · 24 context-only

  • Instruction Following5%

    Gray Swan IPI (15 attempts) · IFBench

    2 scored · 15 context-only

  • MultimodalSeparate

    ParseBench-2 · ParseBench-1 · ExtractBench (LlamaIndex) · ExtractBench Pro · MMMU · MMMU-Pro · OfficeQA Pro · OmniDocBench 1.5 · RealWorldQA · Video-MME (with subtitle) · Video-MME (w/o subtitle) · MathVision · We-Math · DynaMath · MMLongBench-Doc · CC-OCR · AI2D_TEST · CountBench · RefCOCO (avg) · ERQA · VideoMMMU · MLVU (M-Avg) · LVBench · ScreenSpot Pro · MedXpertQA (MM) · ZeroBench · SimpleVQA · V* · CharXiv · CharXiv w/o tools · BabyVision · Blueprint-Bench 2

    32 scored · 47 context-only

  • MultilingualSeparate

    MMLU-ProX · NOVA-63 · INCLUDE · PolyMath · MAXIFE · SWE Multilingual

    6 scored · 9 context-only

Scoring methodology · BenchAlign v6.0

Inspect method