Best AI models for deep agentic chains
Ranked by Effective Value at task depth N=40 β long agentic chains where one error compounds. As depth grows the ranking shifts decisively toward accuracy: fast-but-flawed models collapse while accurate models pull ahead. This is the leaderboard for autonomous, multi-step work. Shown indexed to the top model at N=40 = 100.
Ranking 22 models across the meo-tested roster Β· as of 2026-06-08.
| # | Model | Lab | π idx (N=40) |
|---|---|---|---|
| 1 | OpenAI: GPT-5.5 | openai | 100.0 |
| 2 | Anthropic: Claude Opus 4.8 | anthropic | 37.2 |
| 3 | Google: Gemini 3.5 Flash | 0.011 | |
| 4 | Google: Gemini 3.1 Pro Preview | 0.007 | |
| 5 | xAI: Grok 4.3 | x-ai | 0.002 |
| 6 | inclusionAI: Ring-2.6-1T | inclusionai | <0.001 |
| 7 | Qwen: Qwen3.7 Max | qwen | <0.001 |
| 8 | DeepSeek: DeepSeek V4 Flash | deepseek | <0.001 |
| 9 | MoonshotAI: Kimi K2.6 | moonshotai | <0.001 |
| 10 | NVIDIA: Nemotron 3 Ultra | nvidia | <0.001 |
| 11 | Xiaomi: MiMo-V2.5 | xiaomi | <0.001 |
| 12 | DeepSeek: DeepSeek V4 Pro | deepseek | <0.001 |
| 13 | MiniMax: MiniMax M3 | minimax | <0.001 |
| 14 | Qwen: Qwen3.7 Plus | qwen | <0.001 |
| 15 | Xiaomi: MiMo-V2.5-Pro | xiaomi | <0.001 |
| 16 | Perceptron: Perceptron Mk1 | perceptron | <0.001 |
| 17 | Owl Alpha | openrouter | <0.001 |
| 18 | Google: Gemini 3.1 Flash Lite | <0.001 | |
| 19 | Z.ai: GLM 5.1 | z-ai | <0.001 |
| 20 | StepFun: Step 3.7 Flash | stepfun | <0.001 |
| 21 | Tencent: Hy3 preview | tencent | <0.001 |
| 22 | Arcee AI: Trinity Large Thinking | arcee-ai | <0.001 |