DeepSeek: DeepSeek V3.2: DeepSeek-V3.2 is a large language model designed to harmonize high computational efficiency with strong reasoning and agentic tool-use performance. It introduces DeepSeek Sparse Attention (DSA), a fine-grained sparse attention mechanism...
No signals on file for this pillar yet — it contributes 0 to the headline. The headline is a flat weighted sum, so a missing pillar costs its full weight (no redistribution).
Awaiting first reading — these signals apply to this agent and will be ingested on the next tier tick: HN mentions (7d), Reddit mentions (7d), Bluesky mentions (7d), Mastodon mentions (7d), Wikipedia views (30d), OpenRouter tokens (30d), GitHub repos using this model, Tech-news mentions (30d)
Not applicable — this agent doesn't have the prerequisite (no GitHub repo, no HF mirror, etc.) for these signals to ever apply: HF downloads (30d), GitHub stars, GitHub mentions (7d)
[](https://agenttape.com/agents/deepseek-deepseek-v3-2)
<a href="https://agenttape.com/agents/deepseek-deepseek-v3-2"><img src="https://agenttape.com/api/badge/deepseek-deepseek-v3-2.svg" alt="AgentTape" /></a>
| Benchmark | Score | Max | Captured |
|---|---|---|---|
| aa-coding-index | 34.60 | 100.00 | May 18, 2026 |
| aa-intelligence-index | 24.70 | 100.00 | 8d ago |
| aa-math-index | 59.00 | 100.00 | May 18, 2026 |
| gpqa-diamond | 75.10 | 100.00 | May 18, 2026 |
| ifbench | 49.05 | 100.00 | 10h ago |
| livecodebench | 59.30 | 100.00 | May 18, 2026 |
| lmarena | 1387.24 | 1500.00 | 19d ago |
| mmlu-pro |
| 83.70 |
| 100.00 |
| May 18, 2026 |
| scicode | 38.70 | 100.00 | May 18, 2026 |