---
title: "Agents leaderboard"
description: "Rankings on finishing multi-step tasks in real agent sessions; scores are net improvement over a baseline (percentage points)."
canonical: https://pirouter.ai/rankings/agent
last-updated: 2026-10-15
---

# Agents leaderboard

> Rankings on finishing multi-step tasks in real agent sessions; scores are net improvement over a baseline (percentage points).

- Updated: 2026-09-30
- 1,897,511 sessions · 44 models
- Net improvement (pp) · higher is better

## Full ranking

| Rank | Rank spread | Model | Vendor | Score | Votes |
|---|---|---|---|---|---|
| 1 | 1–4 | [Claude Fable 5.1 \(max\)](https://pirouter.ai/models/anthropic/claude-fable-5.1) | Anthropic | +14.6 ±2.0 | 14636 |
| 2 | 1–4 | [Claude Opus 5.5 \(high\)](https://pirouter.ai/models/anthropic/claude-opus-5.5) | Anthropic | +13.8 ±2.3 | 5013 |
| 3 | 1–6 | [GPT-6 Astra \(max\)](https://pirouter.ai/models/openai/gpt-6-astra) | OpenAI | +12.2 ±2.3 | 11077 |
| 4 | 1–9 | [GPT-6 Sol \(max\)](https://pirouter.ai/models/openai/gpt-6-sol) | OpenAI | +10.6 ±2.5 | 5031 |
| 5 | 3–9 | Claude Opus 5 (high) | Anthropic | +8.8 ±1.5 | 26501 |
| 6 | 4–9 | Claude Fable 5 (high) | Anthropic | +8.4 ±1.3 | 39998 |
| 7 | 3–14 | Gemini 4 Argon (high) | Google | +7.9 ±3.2 | 3417 |
| 8 | 4–10 | [GPT-5.6 Sol \(xhigh\)](https://pirouter.ai/models/openai/gpt-5.6-sol) | OpenAI | +7.1 ±1.4 | 33916 |
| 9 | 4–10 | Claude Opus 4.8 (high) | Anthropic | +6.9 ±1.4 | 40210 |
| 10 | 7–21 | [Claude Sonnet 5 \(high\)](https://pirouter.ai/models/anthropic/claude-sonnet-5) | Anthropic | +4.8 ±1.7 | 31480 |
| 11 | 9–19 | Hy4 Preview | Tencent | +4.2 ±0.8 | 53732 |
| 12 | 9–18 | Muse Spark 1.3 (max) | Meta | +4.2 ±0.6 | 61422 |
| 13 | 9–21 | [GPT-5.5 \(xhigh\)](https://pirouter.ai/models/openai/gpt-5.5) | OpenAI | +4.1 ±1.1 | 53913 |
| 14 | 10–18 | [DeepSeek V4.1 Flash \(max\)](https://pirouter.ai/models/deepseek/deepseek-v4.1-flash) | DeepSeek | +4.0 ±0.5 | 95889 |
| 15 | 10–19 | Kimi K3 (max) | Moonshot | +4.0 ±0.5 | 129908 |
| 16 | 9–25 | Grok 4.7 (xhigh) | xAI | +3.6 ±1.6 | 10109 |
| 17 | 10–22 | [GLM 5.2 \(max\)](https://pirouter.ai/models/z-ai/glm-5.2) | Z.ai | +3.6 ±0.8 | 83747 |
| 18 | 10–23 | [Gemini 3.8 Flash \(high\)](https://pirouter.ai/models/google/gemini-3.8-flash) | Google | +3.0 ±0.8 | 32890 |
| 19 | 12–23 | Qwen3.8 (max) | Alibaba | +2.8 ±0.6 | 49157 |
| 20 | 14–25 | GLM 5.3 (max) | Z.ai | +2.6 ±0.7 | 80704 |
| 21 | 14–26 | [GPT-6 Luna \(max\)](https://pirouter.ai/models/openai/gpt-6-luna) | OpenAI | +1.7 ±1.4 | 12076 |
| 22 | 16–29 | [DeepSeek V4 Pro 0813 \(high\)](https://pirouter.ai/models/deepseek/deepseek-v4-pro) | DeepSeek | +1.3 ±1.7 | 6743 |
| 23 | 17–27 | Grok 4.6 (xhigh) | xAI | +1.2 ±1.1 | 25563 |
| 24 | 19–27 | Grok 4.5 | xAI | +1.1 ±1.0 | 41479 |
| 25 | 21–27 | [GPT-5.4 \(high\)](https://pirouter.ai/models/openai/gpt-5.4) | OpenAI | +0.9 ±0.9 | 81278 |
| 26 | 19–29 | [GPT-5.6 Terra \(xhigh\)](https://pirouter.ai/models/openai/gpt-5.6-terra) | OpenAI | +0.8 ±1.3 | 22046 |
| 27 | 22–29 | [GLM 5.3 Flash](https://pirouter.ai/models/z-ai/glm-5.3-flash) | Z.ai | -0.3 ±0.5 | 76297 |
| 28 | 25–31 | Qwen3.8 Flash Next | Alibaba | -0.8 ±0.6 | 78117 |
| 29 | 25–31 | [GPT-5.6 Luna \(xhigh\)](https://pirouter.ai/models/openai/gpt-5.6-luna) | OpenAI | -1.1 ±0.7 | 39830 |
| 30 | 28–31 | Qwen 3.8 27b | Alibaba | -1.6 ±0.6 | 53985 |
| 31 | 28–32 | [Gemini 3.7 Flash \(high\)](https://pirouter.ai/models/google/gemini-3.7-flash) | Google | -1.6 ±0.8 | 51614 |
| 32 | 31–32 | Muse Spark 1.2 (xhigh) | Meta | -3.0 ±0.7 | 46284 |
| 33 | 33–35 | Muse Spark 1.1 | Meta | -4.6 ±0.5 | 139186 |
| 34 | 33–37 | Qwen3.7 (max) | Alibaba | -5.2 ±0.9 | 39957 |
| 35 | 33–40 | Hy3 | Tencent | -5.7 ±1.2 | 29011 |
| 36 | 34–40 | Minimax M3 | MiniMax | -6.7 ±0.9 | 40771 |
| 37 | 35–40 | MiMo V2.5 Pro | Xiaomi | -7.2 ±0.9 | 44405 |
| 38 | 34–41 | Qwen3.7 Plus | Alibaba | -7.3 ±1.4 | 21406 |
| 39 | 35–40 | Gemini 3.1 Pro Preview | Google | -7.3 ±1.0 | 86482 |
| 40 | 35–41 | [Gemini 3.6 Flash \(high\)](https://pirouter.ai/models/google/gemini-3.6-flash) | Google | -7.7 ±1.1 | 22568 |
| 41 | 39–43 | Inkling Small | Thinky | -9.8 ±1.4 | 13028 |
| 42 | 41–43 | Inkling | Thinky | -10.9 ±1.1 | 42596 |
| 43 | 41–43 | Mistral Medium 3.5 | Mistral | -11.7 ±1.5 | 10797 |
| 44 | 44–44 | Solar Pro 4 | Upstage | -15.8 ±1.9 | 9242 |

## FAQ

### What is the best AI model for agents?

As of 2026-09-30, Anthropic Claude Fable 5.1 (max) is #1 on the agents board with a net improvement (pp) of +14.6, rank spread 1–4.

### What is the best open-weights model for agents?

Hy4 Preview (Tencent), #11 on this board.

### What does this board rank?

Rankings on finishing multi-step tasks in real agent sessions; scores are net improvement over a baseline (percentage points).

### How often is it updated?

Whenever new models ship or the rankings change; this board was last updated on 2026-09-30.

Data: [LMArena leaderboard dataset](https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset), used under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/); filtered and reasoning tiers merged, scores unchanged.
View the original board: https://arena.ai/leaderboard/agent