benchmarks
This profile is specific to DeepSeek V4 Pro 0423; findings about other DeepSeek variants do not establish its capabilities or vulnerabilities. Not yet confirmed by public reporting.
Benchmark results
Every figure links to where it was published. Compare like with like — harnesses and prompt setups differ between vendors.
| Benchmark | Score | Source |
|---|---|---|
| GPQA DiamondPhD-level science questions | 91.7%18 Aug 2026 | View |
| SWE-bench VerifiedResolving real GitHub issues | 77.6%18 Jun 2026 | View |
| OTIS Mock AIME 2024-25Competition mathematics | 98.6%18 Aug 2026 | View |
| SimpleQA VerifiedShort factual questions (hallucination resistance) | 52.9%27 Aug 2026 | View |
| LMArena TextHuman preference Elo from blind chat votes (not a percentage) | 1451 Elo22 Oct 2026 | View |
| LMArena WebDevHuman preference Elo for building web apps (not a percentage) | 1464 Elo23 Oct 2026 | View |
Weighted mean of 4 verified benchmark results: GPQA Diamond 91.7%, SWE-bench Verified 77.6%, OTIS Mock AIME 2024-25 98.6%, SimpleQA Verified 52.9%.
Risks, safety & security issues
- A reported DeepSeek Harness flaw allowed an agent to disable its own file sandbox; this concerns the execution tool rather than an established model vulnerability.
- Anthropic reportedly accused DeepSeek and other laboratories of illicit Claude distillation; the allegations concern organizational activity rather than a demonstrated V4 Pro 0423 weakness.
Profile last reviewed 10/27/2026 and re-checked automatically every few hours against the vendor's published documentation, papers and public leaderboards. Scores come from the cited benchmark results above — confirm limits and pricing against the provider's own documentation before building on them.