BlogLLM Benchmark Scores as Published, Side by Side
DeepSeek-V4-Pro-0813: Benchmark Scores as Published
Every result the catalog holds for DeepSeek-V4-Pro-0813: 30 tests in 8 areas, from 6 pages. 19 printed by DeepSeek itself, the rest by the company named beside each number. Each number is as printed; nothing is ranked or rescored.
Developer DeepSeekInput TextOutput TextWeights Open (MIT)Data version 2026103001, as of 2026-10-30
Knowledge Work
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| AutomationBench | 31.8% | DeepSeek, DeepSeek-V4-Pro-0813 model cardchecked 2026-10-15; printed on 3 more pages | - |
| GDPval-AA v2 | 1590 | Z.ai, GLM-5.3 model cardchecked 2026-10-15 | - |
| Vals Finance Agent v2 | 50.4% | Mistral AI, Mistral Large 4 (preview)checked 2026-10-15 | - |
| Harvey's Legal Agent Benchmark | 7.5% | Mistral AI, Mistral Large 4 (preview)checked 2026-10-15 | - |
| AA-Briefcase | 1256 | Mistral AI, Mistral Large 4 (preview)checked 2026-10-15 | - |
| FinWorkBench (Finch) | 67.4% | Mistral AI, Mistral Large 4 (preview)checked 2026-10-15 | - |
Agentic Coding
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Terminal-Bench 2.1 | 87.9% | DeepSeek, DeepSeek-V4-Pro-0813 model cardchecked 2026-10-15; printed on 2 more pages | - |
| DeepSWE v1.1 | 62.7% | DeepSeek, DeepSeek-V4-Pro-0813 model cardchecked 2026-10-15; printed on 3 more pages | - |
| Terminal-Bench 4.0 | 12.4% | DeepSeek, DeepSeek-V4.1-Flash model cardchecked 2026-10-15; printed on 1 more page | Effort: max |
| NL2Repo-Bench | 61.5 | DeepSeek, DeepSeek-V4-Pro-0813 model cardchecked 2026-10-15; printed on 2 more pages | - |
| ProgramBench Almost@1 | 15.5% | DeepSeek, DeepSeek-V4.1-Flash model cardchecked 2026-10-15 | Effort: max |
| Terminal-Bench 3.0 | 11.8% | DeepSeek, DeepSeek-V4.1-Flash model cardchecked 2026-10-15 | Effort: max |
| SWE-Atlas-QnA (setting not stated) | 66% | Mistral AI, Mistral Large 4 (preview)checked 2026-10-15 | - |
| DSBench-FullStack | 71.1% | DeepSeek, DeepSeek-V4-Pro-0813 model cardchecked 2026-10-15 | - |
| DSBench-Hard | 67.2% | DeepSeek, DeepSeek-V4-Pro-0813 model cardchecked 2026-10-15 | - |
Science and Math
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPQA Diamond | 92.4% | DeepSeek, DeepSeek-V4.1-Flash model cardchecked 2026-10-15 | Effort: max |
| MathArena Apex | 65.3% | DeepSeek, DeepSeek-V4.1-Flash model cardchecked 2026-10-15 | Effort: max |
| SciCode-Verified | 91% | Mistral AI, Mistral Large 4 (preview)checked 2026-10-15 | Attempt budget: pass@1 (n=6) |
Cybersecurity
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| CyberGym | 83.3% | DeepSeek, DeepSeek-V4-Pro-0813 model cardchecked 2026-10-15; printed on 2 more pages | - |
| ExploitGym | 5.4% | DeepSeek, DeepSeek-V4.1-Flash model cardchecked 2026-10-15 | Effort: max |
| SEC-Bench Pro | 56.4% | DeepSeek, DeepSeek-V4.1-Flash model cardchecked 2026-10-15 | Effort: max |
| Cybench | 88% | Mistral AI, Mistral Large 4 (preview)checked 2026-10-15 | - |
General Knowledge and Reasoning
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Humanity's Last Exam (no tools) | 42.7% | DeepSeek, DeepSeek-V4-Pro-0813 model cardchecked 2026-10-15; printed on 1 more page | Tools: none |
| Humanity's Last Exam (with tools) | 60% | DeepSeek, DeepSeek-V4-Pro-0813 model cardchecked 2026-10-15; printed on 2 more pages | Tools: enabled; exact tools in source |
Code Generation
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Codeforces rating | 3348 | DeepSeek, DeepSeek-V4.1-Flash model cardchecked 2026-10-15 | Effort: max |
Tool Use and Workflows
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Agents' Last Exam (pass rate) | 25.7% | DeepSeek, DeepSeek-V4-Pro-0813 model cardchecked 2026-10-15; printed on 2 more pages | - |
| Toolathlon Verified | 74.1% | DeepSeek, DeepSeek-V4-Pro-0813 model cardchecked 2026-10-15; printed on 1 more page | - |
Safety and Alignment
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| B3 AI Security Benchmark (attack resistance) | 85.2% | Mistral AI, Mistral Large 4 (preview)checked 2026-10-15 | - |
| Cyber refusal rate (average of JailbreakBench, StrongREJECT, AgentHarm) | 77.7% | Mistral AI, Mistral Large 4 (preview)checked 2026-10-15 | - |
| KORA Benchmark (0 to 2) | 1.4 | Mistral AI, Mistral Large 4 (preview)checked 2026-10-15 | - |
Price
API list prices in US dollars per one million tokens, as printed on the developer's pricing page on the day checked. Cost per task is measured by Vals AI.
| Price | Value | Printed by | Setup as printed |
|---|---|---|---|
| Input | $1.32 | DeepSeek, DeepSeek API pricingchecked 2026-10-25 | - |
| Cached input | $0.044 | DeepSeek, DeepSeek API pricingchecked 2026-10-25 | - |
| Output | $3.96 | DeepSeek, DeepSeek API pricingchecked 2026-10-25 | - |
| Input, off-peak hours | $0.66 | DeepSeek, DeepSeek API pricingchecked 2026-10-25 | - |
| Cached input, off-peak hours | $0.022 | DeepSeek, DeepSeek API pricingchecked 2026-10-25 | - |
| Output, off-peak hours | $1.98 | DeepSeek, DeepSeek API pricingchecked 2026-10-25 | - |
| Cost per taskVals AI, USD, lower is better | $0.36 | Vals AI, Vibe Code Bench v1.1 leaderboardchecked 2026-10-27 | Harness: OpenHands |
More from DeepSeek
- DeepSeek-V4.1-Flash-1017
- DeepSeek-V4.1-Flash
- DeepSeek-V4-Flash (Preview)
- DeepSeek-V4-Flash-Base
- DeepSeek-V4-Pro (Preview)
- DeepSeek-V4-Pro-Base
- DeepSeek-V3.2
- DeepSeek-V3.2-Base
- DeepSeek-V3.2-Speciale
- DeepSeek-V3.2-Exp
- DeepSeek-V3.1-Terminus
- DeepSeek-V3.1
- DeepSeek-R1-0528
- DeepSeek-R1-0528-Qwen3-8B
- DeepSeek-V3-0324
- DeepSeek-R1
- DeepSeek-R1-Distill-Llama-70B
- DeepSeek-R1-Distill-Llama-8B
- DeepSeek-R1-Distill-Qwen-1.5B
- DeepSeek-R1-Distill-Qwen-14B
- DeepSeek-R1-Distill-Qwen-32B
- DeepSeek-R1-Distill-Qwen-7B
- DeepSeek-V3
- DeepSeek-V4-Flash-0731
- DeepSeek-V4-Vision-Exp
Every number here is read from the same catalog file as the benchmark table, data version 2026103001. Where one page printed the model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.