# BenchLeader: full summary (data as of 2026-10-25)

> BenchLeader (benchleader.com) is an independent LLM leaderboard that ranks about 450 AI models every day by quality, speed and price, using results for 240 benchmarks from 43 benchmark publishers rather than its own tests or vendor launch claims. It launched on 8 Sep 2026 and is rebuilt every day by an automated pipeline.

Hundreds of AI models ranked daily by quality, speed, latency and price, reconciled from 240 benchmarks from 43 benchmark publishers. Every score links to its source.

Canonical site: https://www.benchleader.com. Updated once a day (the run starts at 22:03 UTC). This file is regenerated with every update; cite the date above.
BenchLeader's own derived figures (the BenchLeader Index, its interval and the category scores) are licensed CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/): non-commercial use only, with attribution — credit "BenchLeader" with a link to https://www.benchleader.com (or to the row's url). That licence does not cover anyone else's figures: list prices are the vendors' own published facts, and speed, latency and the benchmark results behind the scores remain under their publishers' terms (credited on each model page and listed under "sources"; Epoch AI's Benchmarking Hub, for example, is CC BY 4.0).

## Methodology in brief
- BenchLeader is an aggregator: it runs no evaluations. It collects results already published by independent evaluators and arenas (240 benchmarks from 43 benchmark publishers) and reconciles the names they use for the same model. Launch-post scores never enter a ranking or a benchmark board. A few specs (list price, context window, time to first answer) may come from the vendor when no independent source lists the model yet; those are labelled vendor-reported, as are the few maker-run scores shown on a model’s own page.
- Each row is a configuration: one model at one reasoning-effort setting, scored only on results measured at that setting. Rankings below show one row per model, at its best-scoring configuration.
- Every quality benchmark is standardised so that 50 is the average evaluated configuration and 15 points is one standard deviation. Benchmarks are averaged within their category, categories into the BenchLeader Index.
- The index needs at least five results across at least two categories, and is shrunk toward 50 by n/(n+3), n being the number of results behind it. The ± figure is the half-width of an approximate 95% band: how much its results disagree with each other.
- Prices are list prices per million tokens, blended at three input tokens per output token. Output tokens per second: Artificial Analysis' measured median where it covers a model, otherwise OpenRouter's live-traffic median smoothed over the last seven days.
- [Full method](https://www.benchleader.com/methodology)
- [About BenchLeader](https://www.benchleader.com/about): how it is run, contact and corrections

## How to cite
Cite the page you used, not the site: "BenchLeader", the page title, the data date shown on the page ("data as of"), and the page's canonical URL. Example: BenchLeader, “<page title>”, data as of <date>, https://www.benchleader.com/<path>.

## Top 20 overall: BenchLeader Index (October 2026)
Source page: [BenchLeader Index](https://www.benchleader.com/)
1. [Claude Opus 5.5 (max effort)](https://www.benchleader.com/models/claude-opus-5-5) by Anthropic: index 72.6 ±5.6; $8.00/M blended; 96 tok/s.
2. [Gemini 4 Argon (high effort)](https://www.benchleader.com/models/gemini-4-argon) by Google: index 70.7 ±2.7; $4.00/M blended; 79 tok/s.
3. [Muse Wonder Full (max effort)](https://www.benchleader.com/models/muse-wonder-full) by Meta: index 70.6 ±8.7; price n/a; speed n/a.
4. [Claude Fable 5.1 (max effort)](https://www.benchleader.com/models/claude-fable-5-1) by Anthropic: index 70.5 ±4.6; $20.00/M blended; 68 tok/s.
5. [GPT-6 Astra (max effort)](https://www.benchleader.com/models/gpt-6-astra) by OpenAI: index 70.4 ±3.7; $20.00/M blended; 58 tok/s.
6. [GPT-6.1 Sol (max effort)](https://www.benchleader.com/models/gpt-6-1-sol) by OpenAI: index 69.9 ±3.3; $4.00/M blended; 57 tok/s.
7. [Claude Sonnet 5.5 (max effort)](https://www.benchleader.com/models/claude-sonnet-5-5) by Anthropic: index 69.4 ±4.9; $4.00/M blended; 133 tok/s.
8. [Claude Fable 5 (max effort)](https://www.benchleader.com/models/claude-fable-5) by Anthropic: index 69.3 ±4.3; $20.00/M blended; 70 tok/s.
9. [Claude Opus 5 (max effort)](https://www.benchleader.com/models/claude-opus-5) by Anthropic: index 68.8 ±2.7; $10.00/M blended; 57 tok/s.
10. [GPT-5.6 Sol (max effort)](https://www.benchleader.com/models/gpt-5-6-sol) by OpenAI: index 68.0 ±3.7; $8.00/M blended; 89 tok/s.
11. [Muse Wonder (max effort)](https://www.benchleader.com/models/muse-wonder) by Meta: index 67.5 ±4.3; $8.00/M blended; 145 tok/s.
12. [Muse Spark 1.3 (max effort)](https://www.benchleader.com/models/muse-spark-1-3) by Meta: index 67.1 ±4.4; $2.00/M blended; 125 tok/s.
13. [Kimi K3.1 Pro](https://www.benchleader.com/models/kimi-k3-1-pro) by Moonshot AI: index 66.9 ±6.2; $4.00/M blended; 73 tok/s.
14. [GPT-6 Sol (max effort)](https://www.benchleader.com/models/gpt-6-sol) by OpenAI: index 66.4 ±3.9; $4.00/M blended; 103 tok/s.
15. [GPT-5.5 (xhigh effort)](https://www.benchleader.com/models/gpt-5-5) by OpenAI: index 66.1 ±3.7; $11.25/M blended; 86 tok/s.
16. [GPT-5.5 Pro (xhigh effort)](https://www.benchleader.com/models/gpt-5-5-pro) by OpenAI: index 65.3 ±8.2; $67.50/M blended; 27 tok/s.
17. [Kimi K3 (max effort)](https://www.benchleader.com/models/kimi-k3) by Moonshot AI: index 64.8 ±3.3; $6.00/M blended; 41 tok/s.
18. [GPT-5.4 (xhigh effort)](https://www.benchleader.com/models/gpt-5-4) by OpenAI: index 64.7 ±2.6; $5.63/M blended; 96 tok/s.
19. [Gemini 3.8 Flash (high effort)](https://www.benchleader.com/models/gemini-3-8-flash) by Google: index 64.4 ±3.9; $1.50/M blended; 141 tok/s.
20. [GPT-5.6 Terra (max effort)](https://www.benchleader.com/models/gpt-5-6-terra) by OpenAI: index 64.4 ±5.0; $4.50/M blended; 126 tok/s.

## Top 20: Reasoning (category score, models with results on at least two of its benchmarks)
1. [Muse Wonder (max effort)](https://www.benchleader.com/models/muse-wonder) by Meta: reasoning 81.3; $8.00/M blended; 145 tok/s.
2. [MiMo-V2.6-Pro](https://www.benchleader.com/models/mimo-v2-6-pro) by Xiaomi: reasoning 78.7; $0.55/M blended; 55 tok/s.
3. [GPT-5.5 Pro (xhigh effort)](https://www.benchleader.com/models/gpt-5-5-pro) by OpenAI: reasoning 77.6; $67.50/M blended; 27 tok/s.
4. [Claude Sonnet 5.5 (max effort)](https://www.benchleader.com/models/claude-sonnet-5-5) by Anthropic: reasoning 77.4; $4.00/M blended; 133 tok/s.
5. [Muse Spark 1.3 (max effort)](https://www.benchleader.com/models/muse-spark-1-3) by Meta: reasoning 77.4; $2.00/M blended; 125 tok/s.
6. [Claude Fable 5.1 (max effort)](https://www.benchleader.com/models/claude-fable-5-1) by Anthropic: reasoning 76.8; $20.00/M blended; 68 tok/s.
7. [GPT-6 Astra (max effort)](https://www.benchleader.com/models/gpt-6-astra) by OpenAI: reasoning 76.2; $20.00/M blended; 58 tok/s.
8. [GPT-6.1 Sol (max effort)](https://www.benchleader.com/models/gpt-6-1-sol) by OpenAI: reasoning 76.0; $4.00/M blended; 57 tok/s.
9. [Claude Opus 5.5 (max effort)](https://www.benchleader.com/models/claude-opus-5-5) by Anthropic: reasoning 75.7; $8.00/M blended; 96 tok/s.
10. [GPT-5.5 (xhigh effort)](https://www.benchleader.com/models/gpt-5-5) by OpenAI: reasoning 75.5; $11.25/M blended; 86 tok/s.
11. [Gemini 4 Argon (high effort)](https://www.benchleader.com/models/gemini-4-argon) by Google: reasoning 74.8; $4.00/M blended; 79 tok/s.
12. [Gemini 3 Deep Think (thinking effort)](https://www.benchleader.com/models/gemini-3-deep-think) by Google: reasoning 74.6; price n/a; speed n/a.
13. [Claude Opus 5 (max effort)](https://www.benchleader.com/models/claude-opus-5) by Anthropic: reasoning 74.1; $10.00/M blended; 57 tok/s.
14. [Claude Fable 5 (max effort)](https://www.benchleader.com/models/claude-fable-5) by Anthropic: reasoning 73.8; $20.00/M blended; 70 tok/s.
15. [GPT-5.4 Pro](https://www.benchleader.com/models/gpt-5-4-pro) by OpenAI: reasoning 72.5; $67.50/M blended; 7 tok/s.
16. [GPT-5.6 Sol (max effort)](https://www.benchleader.com/models/gpt-5-6-sol) by OpenAI: reasoning 72.4; $8.00/M blended; 89 tok/s.
17. [Qwen3.8 Max (max effort)](https://www.benchleader.com/models/qwen3-8-max) by Alibaba: reasoning 71.6; $3.00/M blended; 42 tok/s.
18. [GPT-6 Sol (max effort)](https://www.benchleader.com/models/gpt-6-sol) by OpenAI: reasoning 71.5; $4.00/M blended; 103 tok/s.
19. [Step 5 Preview](https://www.benchleader.com/models/step-5) by StepFun: reasoning 71.5; $1.43/M blended; 229 tok/s.
20. [DeepSeek V4.1 Flash 1017 (max effort)](https://www.benchleader.com/models/deepseek-v4-1-flash-1017) by DeepSeek: reasoning 70.6; $0.53/M blended; 195 tok/s.

## Top 20: Coding (category score, models with results on at least two of its benchmarks)
1. [Claude Opus 5.5 (max effort)](https://www.benchleader.com/models/claude-opus-5-5) by Anthropic: coding 81.9; $8.00/M blended; 96 tok/s.
2. [Claude Fable 5.1 (max effort)](https://www.benchleader.com/models/claude-fable-5-1) by Anthropic: coding 74.9; $20.00/M blended; 68 tok/s.
3. [GPT-6 Astra (max effort)](https://www.benchleader.com/models/gpt-6-astra) by OpenAI: coding 73.9; $20.00/M blended; 58 tok/s.
4. [Claude Fable 5 (max effort)](https://www.benchleader.com/models/claude-fable-5) by Anthropic: coding 73.8; $20.00/M blended; 70 tok/s.
5. [Muse Wonder Full (max effort)](https://www.benchleader.com/models/muse-wonder-full) by Meta: coding 72.1; price n/a; speed n/a.
6. [GPT-6.1 Sol (max effort)](https://www.benchleader.com/models/gpt-6-1-sol) by OpenAI: coding 71.9; $4.00/M blended; 57 tok/s.
7. [Gemini 4 Argon (high effort)](https://www.benchleader.com/models/gemini-4-argon) by Google: coding 71.7; $4.00/M blended; 79 tok/s.
8. [Claude Opus 5 (max effort)](https://www.benchleader.com/models/claude-opus-5) by Anthropic: coding 70.2; $10.00/M blended; 57 tok/s.
9. [GPT-6 Sol (max effort)](https://www.benchleader.com/models/gpt-6-sol) by OpenAI: coding 67.2; $4.00/M blended; 103 tok/s.
10. [Muse Spark 1.1](https://www.benchleader.com/models/muse-spark-1-1) by Meta: coding 66.3; $2.00/M blended; 200 tok/s.
11. [Muse Wonder (max effort)](https://www.benchleader.com/models/muse-wonder) by Meta: coding 66.1; $8.00/M blended; 145 tok/s.
12. [GPT-5.6 Sol (max effort)](https://www.benchleader.com/models/gpt-5-6-sol) by OpenAI: coding 65.9; $8.00/M blended; 89 tok/s.
13. [GPT-5.6 Terra (max effort)](https://www.benchleader.com/models/gpt-5-6-terra) by OpenAI: coding 65.1; $4.50/M blended; 126 tok/s.
14. [Muse Spark](https://www.benchleader.com/models/muse-spark) by Meta: coding 65.0; price n/a; speed n/a.
15. [Kimi K3 (max effort)](https://www.benchleader.com/models/kimi-k3) by Moonshot AI: coding 64.6; $6.00/M blended; 41 tok/s.
16. [GPT-5.4 (xhigh effort)](https://www.benchleader.com/models/gpt-5-4) by OpenAI: coding 64.6; $5.63/M blended; 96 tok/s.
17. [Qwen3.8 Max (max effort)](https://www.benchleader.com/models/qwen3-8-max) by Alibaba: coding 64.5; $3.00/M blended; 42 tok/s.
18. [Muse Spark 1.3 (max effort)](https://www.benchleader.com/models/muse-spark-1-3) by Meta: coding 64.4; $2.00/M blended; 125 tok/s.
19. [Gemini 3.8 Flash (high effort)](https://www.benchleader.com/models/gemini-3-8-flash) by Google: coding 63.4; $1.50/M blended; 141 tok/s.
20. [Gemini 3.7 Flash (high effort)](https://www.benchleader.com/models/gemini-3-7-flash) by Google: coding 62.9; $1.50/M blended; 88 tok/s.

## Top 20: Agents & tools (category score, models with results on at least two of its benchmarks)
1. [Grok 4.1 Fast (thinking effort)](https://www.benchleader.com/models/grok-4-1-fast) by SpaceXAI: agents & tools 70.5; $0.28/M blended; speed n/a.
2. [o3](https://www.benchleader.com/models/o3) by OpenAI: agents & tools 70.3; $3.50/M blended; 134 tok/s.
3. [Claude Sonnet 5.5 (max effort)](https://www.benchleader.com/models/claude-sonnet-5-5) by Anthropic: agents & tools 70.1; $4.00/M blended; 133 tok/s.
4. [Kimi K3.1 Pro](https://www.benchleader.com/models/kimi-k3-1-pro) by Moonshot AI: agents & tools 68.0; $4.00/M blended; 73 tok/s.
5. [GPT-5.6 Sol (max effort)](https://www.benchleader.com/models/gpt-5-6-sol) by OpenAI: agents & tools 66.5; $8.00/M blended; 89 tok/s.
6. [Muse Spark](https://www.benchleader.com/models/muse-spark) by Meta: agents & tools 66.1; price n/a; speed n/a.
7. [Muse Spark 1.3 (max effort)](https://www.benchleader.com/models/muse-spark-1-3) by Meta: agents & tools 65.6; $2.00/M blended; 125 tok/s.
8. [Claude Opus 4.7](https://www.benchleader.com/models/claude-opus-4-7) by Anthropic: agents & tools 65.4; $10.00/M blended; 60 tok/s.
9. [Claude Fable 5 (max effort)](https://www.benchleader.com/models/claude-fable-5) by Anthropic: agents & tools 65.0; $20.00/M blended; 70 tok/s.
10. [GPT-6.1 Sol (max effort)](https://www.benchleader.com/models/gpt-6-1-sol) by OpenAI: agents & tools 64.5; $4.00/M blended; 57 tok/s.
11. [Qwen3.8-Flash-Next](https://www.benchleader.com/models/qwen3-8-flash-next) by Alibaba: agents & tools 64.4; $0.23/M blended; 52 tok/s.
12. [Claude Opus 5 (max effort)](https://www.benchleader.com/models/claude-opus-5) by Anthropic: agents & tools 64.3; $10.00/M blended; 57 tok/s.
13. [GPT-6 Astra (max effort)](https://www.benchleader.com/models/gpt-6-astra) by OpenAI: agents & tools 64.2; $20.00/M blended; 58 tok/s.
14. [DeepSeek V4 Pro 0424 (max effort)](https://www.benchleader.com/models/deepseek-v4-pro-0424) by DeepSeek: agents & tools 63.8; $0.55/M blended; 102 tok/s.
15. [GPT-5.4 (xhigh effort)](https://www.benchleader.com/models/gpt-5-4) by OpenAI: agents & tools 63.6; $5.63/M blended; 96 tok/s.
16. [DeepSeek V4.1 Flash 1017 (max effort)](https://www.benchleader.com/models/deepseek-v4-1-flash-1017) by DeepSeek: agents & tools 63.5; $0.53/M blended; 195 tok/s.
17. [Gemini 3.5 Flash (high effort)](https://www.benchleader.com/models/gemini-3-5-flash) by Google: agents & tools 63.4; $3.38/M blended; 222 tok/s.
18. [Claude Opus 5.5 (max effort)](https://www.benchleader.com/models/claude-opus-5-5) by Anthropic: agents & tools 63.2; $8.00/M blended; 96 tok/s.
19. [GLM 5.2 (max effort)](https://www.benchleader.com/models/glm-5-2) by Zhipu AI: agents & tools 62.6; $2.15/M blended; 94 tok/s.
20. [GPT-5.5 (xhigh effort)](https://www.benchleader.com/models/gpt-5-5) by OpenAI: agents & tools 62.3; $11.25/M blended; 86 tok/s.

## Top 20: Maths (category score, models with results on at least two of its benchmarks)
1. [Claude Opus 5.5 (max effort)](https://www.benchleader.com/models/claude-opus-5-5) by Anthropic: maths 74.3; $8.00/M blended; 96 tok/s.
2. [GPT-6.1 Sol (max effort)](https://www.benchleader.com/models/gpt-6-1-sol) by OpenAI: maths 73.0; $4.00/M blended; 57 tok/s.
3. [Claude Fable 5.1 (max effort)](https://www.benchleader.com/models/claude-fable-5-1) by Anthropic: maths 72.9; $20.00/M blended; 68 tok/s.
4. [Claude Sonnet 5.5 (max effort)](https://www.benchleader.com/models/claude-sonnet-5-5) by Anthropic: maths 72.4; $4.00/M blended; 133 tok/s.
5. [Claude Fable 5 (max effort)](https://www.benchleader.com/models/claude-fable-5) by Anthropic: maths 72.4; $20.00/M blended; 70 tok/s.
6. [GPT-6 Astra (max effort)](https://www.benchleader.com/models/gpt-6-astra) by OpenAI: maths 71.8; $20.00/M blended; 58 tok/s.
7. [GPT-5.5 Pro (xhigh effort)](https://www.benchleader.com/models/gpt-5-5-pro) by OpenAI: maths 71.3; $67.50/M blended; 27 tok/s.
8. [GPT-5.6 Sol (max effort)](https://www.benchleader.com/models/gpt-5-6-sol) by OpenAI: maths 71.0; $8.00/M blended; 89 tok/s.
9. [Claude Opus 5 (max effort)](https://www.benchleader.com/models/claude-opus-5) by Anthropic: maths 70.6; $10.00/M blended; 57 tok/s.
10. [o3](https://www.benchleader.com/models/o3) by OpenAI: maths 70.6; $3.50/M blended; 134 tok/s.
11. [Gemini 4 Argon (high effort)](https://www.benchleader.com/models/gemini-4-argon) by Google: maths 70.5; $4.00/M blended; 79 tok/s.
12. [GPT-5.5 (xhigh effort)](https://www.benchleader.com/models/gpt-5-5) by OpenAI: maths 69.7; $11.25/M blended; 86 tok/s.
13. [GPT-6 Sol (max effort)](https://www.benchleader.com/models/gpt-6-sol) by OpenAI: maths 69.2; $4.00/M blended; 103 tok/s.
14. [o4-mini](https://www.benchleader.com/models/o4-mini) by OpenAI: maths 69.2; $1.93/M blended; speed n/a.
15. [Qwen3 235B A22B 2507](https://www.benchleader.com/models/qwen3-235b-a22b-2507) by Alibaba: maths 69.1; $0.37/M blended; 37 tok/s.
16. [GPT-5.6 Terra (max effort)](https://www.benchleader.com/models/gpt-5-6-terra) by OpenAI: maths 68.9; $4.50/M blended; 126 tok/s.
17. [Claude Opus 4.8 (max effort)](https://www.benchleader.com/models/claude-opus-4-8) by Anthropic: maths 68.1; $10.00/M blended; 63 tok/s.
18. [Muse Wonder (max effort)](https://www.benchleader.com/models/muse-wonder) by Meta: maths 67.8; $8.00/M blended; 145 tok/s.
19. [Claude Opus 4.7](https://www.benchleader.com/models/claude-opus-4-7) by Anthropic: maths 66.7; $10.00/M blended; 60 tok/s.
20. [MiMo-V2.6-Pro](https://www.benchleader.com/models/mimo-v2-6-pro) by Xiaomi: maths 66.0; $0.55/M blended; 55 tok/s.

## Top 20: Knowledge (category score, models with results on at least two of its benchmarks)
1. [GPT-6 Astra (max effort)](https://www.benchleader.com/models/gpt-6-astra) by OpenAI: knowledge 79.5; $20.00/M blended; 58 tok/s.
2. [Claude Opus 5.5 (max effort)](https://www.benchleader.com/models/claude-opus-5-5) by Anthropic: knowledge 78.8; $8.00/M blended; 96 tok/s.
3. [GPT-6.1 Sol (max effort)](https://www.benchleader.com/models/gpt-6-1-sol) by OpenAI: knowledge 78.4; $4.00/M blended; 57 tok/s.
4. [Claude Fable 5.1 (max effort)](https://www.benchleader.com/models/claude-fable-5-1) by Anthropic: knowledge 77.5; $20.00/M blended; 68 tok/s.
5. [Gemini 4 Argon (high effort)](https://www.benchleader.com/models/gemini-4-argon) by Google: knowledge 76.6; $4.00/M blended; 79 tok/s.
6. [Claude Opus 5 (max effort)](https://www.benchleader.com/models/claude-opus-5) by Anthropic: knowledge 71.6; $10.00/M blended; 57 tok/s.
7. [Muse Wonder Full (max effort)](https://www.benchleader.com/models/muse-wonder-full) by Meta: knowledge 71.4; price n/a; speed n/a.
8. [GPT-6 Sol (max effort)](https://www.benchleader.com/models/gpt-6-sol) by OpenAI: knowledge 69.7; $4.00/M blended; 103 tok/s.
9. [Muse Spark 1.1](https://www.benchleader.com/models/muse-spark-1-1) by Meta: knowledge 69.2; $2.00/M blended; 200 tok/s.
10. [Gemini 3.8 Flash (high effort)](https://www.benchleader.com/models/gemini-3-8-flash) by Google: knowledge 68.4; $1.50/M blended; 141 tok/s.
11. [Gemini 3.7 Flash (high effort)](https://www.benchleader.com/models/gemini-3-7-flash) by Google: knowledge 68.0; $1.50/M blended; 88 tok/s.
12. [Claude Opus 4.8 (max effort)](https://www.benchleader.com/models/claude-opus-4-8) by Anthropic: knowledge 66.9; $10.00/M blended; 63 tok/s.
13. [Muse Spark 1.2 (xhigh effort)](https://www.benchleader.com/models/muse-spark-1-2) by Meta: knowledge 66.7; $2.00/M blended; 307 tok/s.
14. [GPT-5 Pro](https://www.benchleader.com/models/gpt-5-pro) by OpenAI: knowledge 65.8; $41.25/M blended; speed n/a.
15. [Gemini 3.6 Flash (high effort)](https://www.benchleader.com/models/gemini-3-6-flash) by Google: knowledge 65.6; $1.50/M blended; 202 tok/s.
16. [GPT-5.6 Sol (max effort)](https://www.benchleader.com/models/gpt-5-6-sol) by OpenAI: knowledge 65.3; $8.00/M blended; 89 tok/s.
17. [GPT-5.5 (xhigh effort)](https://www.benchleader.com/models/gpt-5-5) by OpenAI: knowledge 65.3; $11.25/M blended; 86 tok/s.
18. [Gemini 3.5 Flash (high effort)](https://www.benchleader.com/models/gemini-3-5-flash) by Google: knowledge 65.2; $3.38/M blended; 222 tok/s.
19. [Claude Sonnet 5.5 (max effort)](https://www.benchleader.com/models/claude-sonnet-5-5) by Anthropic: knowledge 65.0; $4.00/M blended; 133 tok/s.
20. [Gemini 3.1 Pro](https://www.benchleader.com/models/gemini-3-1-pro) by Google: knowledge 63.9; $4.50/M blended; 132 tok/s.

## Top 20: Instruction following (category score, models with results on at least two of its benchmarks)
1. [Muse Spark](https://www.benchleader.com/models/muse-spark) by Meta: instruction following 77.3; price n/a; speed n/a.
2. [o4-mini](https://www.benchleader.com/models/o4-mini) by OpenAI: instruction following 69.0; $1.93/M blended; speed n/a.
3. [GPT-5 mini](https://www.benchleader.com/models/gpt-5-mini) by OpenAI: instruction following 68.8; $0.69/M blended; 87 tok/s.
4. [Gemini 3.1 Pro](https://www.benchleader.com/models/gemini-3-1-pro) by Google: instruction following 67.5; $4.50/M blended; 132 tok/s.
5. [GPT-5 nano](https://www.benchleader.com/models/gpt-5-nano) by OpenAI: instruction following 66.2; $0.14/M blended; 82 tok/s.
6. [o3](https://www.benchleader.com/models/o3) by OpenAI: instruction following 65.1; $3.50/M blended; 134 tok/s.
7. [GPT-5.4 Pro](https://www.benchleader.com/models/gpt-5-4-pro) by OpenAI: instruction following 63.8; $67.50/M blended; 7 tok/s.
8. [Gemini 2.5 Flash](https://www.benchleader.com/models/gemini-2-5-flash) by Google: instruction following 62.6; $0.85/M blended; speed n/a.
9. [Grok 4](https://www.benchleader.com/models/grok-4) by SpaceXAI: instruction following 62.5; $6.00/M blended; 78 tok/s.
10. [Gemini 3.1 Flash Lite](https://www.benchleader.com/models/gemini-3-1-flash-lite) by Google: instruction following 61.2; $0.56/M blended; 390 tok/s.
11. [Gemini 3 Pro](https://www.benchleader.com/models/gemini-3-pro) by Google: instruction following 60.4; price n/a; speed n/a.
12. [Palmyra X 004](https://www.benchleader.com/models/palmyra-x-004) by Writer: instruction following 58.3; price n/a; speed n/a.
13. [Grok 3](https://www.benchleader.com/models/grok-3) by SpaceXAI: instruction following 57.3; $6.00/M blended; 88 tok/s.
14. [Kimi K2 Thinking (thinking effort)](https://www.benchleader.com/models/kimi-k2-thinking) by Moonshot AI: instruction following 56.5; $1.07/M blended; 121 tok/s.
15. [Granite 4.0 Small](https://www.benchleader.com/models/granite-4-0-small) by IBM: instruction following 56.4; price n/a; speed n/a.
16. [Claude 3.5 Sonnet](https://www.benchleader.com/models/claude-3-5-sonnet) by Anthropic: instruction following 55.6; $6.00/M blended; speed n/a.
17. [Qwen3-Next 80B A3B (thinking effort)](https://www.benchleader.com/models/qwen3-next-80b-a3b) by Alibaba: instruction following 54.8; $0.35/M blended; 96 tok/s.
18. [Gemini 1.5 Pro 002](https://www.benchleader.com/models/gemini-1-5-pro-002) by Google: instruction following 54.6; price n/a; speed n/a.
19. [Claude Opus 4 (thinking effort)](https://www.benchleader.com/models/claude-opus-4) by Anthropic: instruction following 53.9; $30.00/M blended; speed n/a.
20. [Qwen3 235B A22B 2507](https://www.benchleader.com/models/qwen3-235b-a22b-2507) by Alibaba: instruction following 53.9; $0.37/M blended; 37 tok/s.

## Top 7: Human preference (category score, models with results on at least two of its benchmarks)
1. [Claude Opus 4.7](https://www.benchleader.com/models/claude-opus-4-7) by Anthropic: human preference 67.5; $10.00/M blended; 60 tok/s.
2. [Muse Spark 1.1](https://www.benchleader.com/models/muse-spark-1-1) by Meta: human preference 65.0; $2.00/M blended; 200 tok/s.
3. [Claude Sonnet 4.6](https://www.benchleader.com/models/claude-sonnet-4-6) by Anthropic: human preference 60.8; $6.00/M blended; 44 tok/s.
4. [Gemini 3.1 Pro](https://www.benchleader.com/models/gemini-3-1-pro) by Google: human preference 58.7; $4.50/M blended; 132 tok/s.
5. [Qwen3.7 (max effort)](https://www.benchleader.com/models/qwen3-7) by Alibaba: human preference 56.0; $3.75/M blended; 201 tok/s.
6. [MiniMax M3](https://www.benchleader.com/models/minimax-m3) by MiniMax: human preference 55.3; $0.53/M blended; 89 tok/s.
7. [Mistral Medium 3.5](https://www.benchleader.com/models/mistral-medium-3-5) by Mistral AI: human preference 46.1; $3.00/M blended; 159 tok/s.

## Top 20: Multimodal (category score, models with results on at least two of its benchmarks)
1. [Gemini 3.8 Flash (high effort)](https://www.benchleader.com/models/gemini-3-8-flash) by Google: multimodal 67.7; $1.50/M blended; 141 tok/s.
2. [Gemini 3.7 Flash (high effort)](https://www.benchleader.com/models/gemini-3-7-flash) by Google: multimodal 67.6; $1.50/M blended; 88 tok/s.
3. [Gemini 3.5 Flash (high effort)](https://www.benchleader.com/models/gemini-3-5-flash) by Google: multimodal 66.6; $3.38/M blended; 222 tok/s.
4. [GPT-6.1 Sol (max effort)](https://www.benchleader.com/models/gpt-6-1-sol) by OpenAI: multimodal 66.1; $4.00/M blended; 57 tok/s.
5. [Qwen3.8 Max (max effort)](https://www.benchleader.com/models/qwen3-8-max) by Alibaba: multimodal 66.1; $3.00/M blended; 42 tok/s.
6. [GPT-6 Astra (max effort)](https://www.benchleader.com/models/gpt-6-astra) by OpenAI: multimodal 66.0; $20.00/M blended; 58 tok/s.
7. [Gemini 3 Pro](https://www.benchleader.com/models/gemini-3-pro) by Google: multimodal 65.6; price n/a; speed n/a.
8. [Gemini 3.6 Flash (high effort)](https://www.benchleader.com/models/gemini-3-6-flash) by Google: multimodal 65.4; $1.50/M blended; 202 tok/s.
9. [Gemini 3.1 Pro](https://www.benchleader.com/models/gemini-3-1-pro) by Google: multimodal 64.8; $4.50/M blended; 132 tok/s.
10. [Muse Spark](https://www.benchleader.com/models/muse-spark) by Meta: multimodal 64.5; price n/a; speed n/a.
11. [Gemini 2.5 Pro](https://www.benchleader.com/models/gemini-2-5-pro) by Google: multimodal 63.6; $3.44/M blended; 121 tok/s.
12. [Qwen3.7 Plus](https://www.benchleader.com/models/qwen3-7-plus) by Alibaba: multimodal 62.6; $0.70/M blended; 59 tok/s.
13. [Gemini 2.5 Flash](https://www.benchleader.com/models/gemini-2-5-flash) by Google: multimodal 61.8; $0.85/M blended; speed n/a.
14. [GPT-6 Sol (max effort)](https://www.benchleader.com/models/gpt-6-sol) by OpenAI: multimodal 61.4; $4.00/M blended; 103 tok/s.
15. [Gemini 3.5 Flash Lite](https://www.benchleader.com/models/gemini-3-5-flash-lite) by Google: multimodal 61.3; $0.85/M blended; 430 tok/s.
16. [Mistral Large 4](https://www.benchleader.com/models/mistral-large-4) by Mistral AI: multimodal 60.2; $2.06/M blended; 95 tok/s.
17. [MiniMax M3](https://www.benchleader.com/models/minimax-m3) by MiniMax: multimodal 59.9; $0.53/M blended; 89 tok/s.
18. [Step 5 Preview](https://www.benchleader.com/models/step-5) by StepFun: multimodal 59.7; $1.43/M blended; 229 tok/s.
19. [Kimi K2.5 (thinking effort)](https://www.benchleader.com/models/kimi-k2-5) by Moonshot AI: multimodal 59.5; $1.20/M blended; 68 tok/s.
20. [Grok 4.20 (thinking effort)](https://www.benchleader.com/models/grok-4-20) by SpaceXAI: multimodal 58.6; $1.56/M blended; 112 tok/s.

## Top 10: Long context (category score, models with results on at least two of its benchmarks)
1. [Grok 4](https://www.benchleader.com/models/grok-4) by SpaceXAI: long context 68.3; $6.00/M blended; 78 tok/s.
2. [Gemini 2.5 Pro](https://www.benchleader.com/models/gemini-2-5-pro) by Google: long context 66.4; $3.44/M blended; 121 tok/s.
3. [GPT-4.1](https://www.benchleader.com/models/gpt-4-1) by OpenAI: long context 56.5; $3.50/M blended; 161 tok/s.
4. [MiniMax-M1](https://www.benchleader.com/models/minimax-m1) by MiniMax: long context 52.7; $0.96/M blended; 95 tok/s.
5. [Grok 3](https://www.benchleader.com/models/grok-3) by SpaceXAI: long context 52.4; $6.00/M blended; 88 tok/s.
6. [Gemini 2.0 Flash](https://www.benchleader.com/models/gemini-2-0-flash) by Google: long context 46.9; price n/a; speed n/a.
7. [Kimi K2 0905](https://www.benchleader.com/models/kimi-k2-0905) by Moonshot AI: long context 45.0; $1.07/M blended; 48 tok/s.
8. [Llama 4 Maverick](https://www.benchleader.com/models/llama-4-maverick) by Meta: long context 42.8; $0.42/M blended; 52 tok/s.
9. [Llama 4 Scout](https://www.benchleader.com/models/llama-4-scout) by Meta: long context 33.9; $0.30/M blended; 103 tok/s.
10. [GPT-4.1 nano](https://www.benchleader.com/models/gpt-4-1-nano) by OpenAI: long context 29.0; $0.17/M blended; 223 tok/s.

## Top 10: Composite (category score, models with results on at least two of its benchmarks)
1. [Kimi K3.1 Pro](https://www.benchleader.com/models/kimi-k3-1-pro) by Moonshot AI: composite 75.8; $4.00/M blended; 73 tok/s.
2. [MiMo-V2.6-Pro](https://www.benchleader.com/models/mimo-v2-6-pro) by Xiaomi: composite 69.8; $0.55/M blended; 55 tok/s.
3. [MiniMax M3.1 Flash](https://www.benchleader.com/models/minimax-m3-1-flash) by MiniMax: composite 66.8; $0.26/M blended; 96 tok/s.
4. [GLM 5.3 Flash](https://www.benchleader.com/models/glm-5-3-flash) by Zhipu AI: composite 64.9; $0.24/M blended; 69 tok/s.
5. [Qwen3.8 Max (max effort)](https://www.benchleader.com/models/qwen3-8-max) by Alibaba: composite 63.3; $3.00/M blended; 42 tok/s.
6. [Step 5 Preview](https://www.benchleader.com/models/step-5) by StepFun: composite 61.1; $1.43/M blended; 229 tok/s.
7. [Mistral Large 4](https://www.benchleader.com/models/mistral-large-4) by Mistral AI: composite 49.9; $2.06/M blended; 95 tok/s.
8. [Beam 1.0](https://www.benchleader.com/models/reflection-beam) by Reflection: composite 49.9; $0.60/M blended; 337 tok/s.
9. [Gemini 3.1 Pro](https://www.benchleader.com/models/gemini-3-1-pro) by Google: composite 49.2; $4.50/M blended; 132 tok/s.
10. [MiniMax M3](https://www.benchleader.com/models/minimax-m3) by MiniMax: composite 48.2; $0.53/M blended; 89 tok/s.

## Best LLMs this month (October 2026)
- [Best LLM overall (October 2026)](https://www.benchleader.com/best/llm): Claude Opus 5.5 (max effort) by Anthropic (BenchLeader Index 72.6)
- [Best LLM for coding (October 2026)](https://www.benchleader.com/best/llm-for-coding): Claude Opus 5.5 (max effort) by Anthropic (coding score 81.9)
- [Best LLM for agents (October 2026)](https://www.benchleader.com/best/llm-for-agents): Claude Sonnet 5.5 (max effort) by Anthropic (agents score 70.1)
- [Best cheap LLM (October 2026)](https://www.benchleader.com/best/cheap-llm): Gemini 3.8 Flash (high effort) by Google (index 64.4 at $1.50/M)
- [Best open-weights LLM (October 2026)](https://www.benchleader.com/best/open-weights-llm): Kimi K3 (max effort) by Moonshot AI (BenchLeader Index 64.8)
- [Fastest LLM (October 2026)](https://www.benchleader.com/best/fastest-llm): K2 Horizon 375B A23B by MBZUAI (output speed 909 tok/s)

## Best model per use case (weighting stated on each page)
- [Voice agents](https://www.benchleader.com/use/voice): DeepSeek V4.1 Flash (high effort) by DeepSeek
- [Coding agents](https://www.benchleader.com/use/coding-agents): DeepSeek V4.1 Flash 1017 (max effort) by DeepSeek
- [High-volume extraction](https://www.benchleader.com/use/bulk-extraction): Ling 3.0 Flash by Ant Group
- [Deep research](https://www.benchleader.com/use/deep-research): Claude Opus 5.5 (xhigh effort) by Anthropic

## More
- [Compare two models](https://www.benchleader.com/compare): /compare/<model-a>-vs-<model-b>, ids in alphabetical order; the reverse order redirects
- [LLM API pricing tracker](https://www.benchleader.com/leaderboards/cost)
- [Reasoning effort compared](https://www.benchleader.com/effort)
- [Leaderboard data (JSON)](https://www.benchleader.com/data/leaderboard.json)
- [Leaderboard data (CSV)](https://www.benchleader.com/data/leaderboard.csv)
- [Short index of the site](https://www.benchleader.com/llms.txt)