Opper TaskBench: leaderboard
Practical AI application tasks including context-grounded QA, natural-language-to-SQL, multi-step agent reasoning with tool selection, and structured data extraction.
Metric: Avg Task Score (%). Source: opper.ai. Saturation forecast: Estimated already saturated. 86 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Fable 5 | 96.4 | #16 |
| 2 | Claude Opus 4.7 | 95.4 | #49 |
| 3 | Gemini 3.1 Pro (Preview) | 95.3 | #58 |
| 4 | Gemini 3 Flash (Preview) | 95.1 | #82 |
| 5 | Gemini 3 Pro (Preview) | 94.8 | #66 |
| 6 | Gemini 3.5 Flash | 94.7 | #60 |
| 7 | Claude Opus 4.8 | 94.5 | #37 |
| 8 | GPT-5.3 Codex | 94.5 | #69 |
| 9 | Claude Opus 4.6 | 94.4 | #63 |
| 10 | Qwen 3.7 Max | 93.3 | #75 |
| 11 | DeepSeek V4 Pro | 93.1 | #102 |
| 12 | Claude Opus 4.5 | 93 | #83 |
| 13 | GLM-5.2 | 93 | #73 |
| 14 | Kimi K2.5 | 92.2 | #144 |
| 15 | GLM-5 | 92.2 | #141 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=opper-taskbench · How It Works · Data refreshed daily, snapshot 2026-10-20.