Context-Bench Filesystem: leaderboard
Letta agentic context engineering benchmark: filesystem suite testing file operations, relationship tracing, and multi-step information retrieval.
Metric: Rubric Score (%). Source: www.letta.com. Source: leaderboard.letta.com. Saturation forecast: Around January 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 Codex (xHigh) | 93 | #95 (GPT-5.2 Codex) |
| 2 | GPT-5.4 (xHigh) | 89 | #79 (GPT-5.4) |
| 3 | Claude Sonnet 4.6 | 88 | #90 |
| 4 | GPT-5.2 (xHigh) | 87 | #110 (GPT-5.2) |
| 5 | GPT-5.3 Codex (xHigh) | 85 | #69 (GPT-5.3 Codex) |
| 6 | Claude Opus 4.6 | 84 | #63 |
| 7 | Gemini 3 Flash | 82 | #99 |
| 8 | GPT-5 Mini (High) | 67 | #186 (GPT-5 Mini) |
| 9 | Claude Opus 4.5 (20251101) | 63 | #74 |
| 10 | Kimi K2.5 | 57 | #144 |
| 11 | Claude Sonnet 4.5 | 48 | #142 |
| 12 | Claude Haiku 4.5 | 43 | #280 |
| 13 | GLM-5 | 39 | #141 |
| 14 | MiniMax-M2.5 | 37 | #303 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=context-bench-filesystem · How It Works · Data refreshed daily, snapshot 2026-10-20.