BlogLLM Benchmark Scores as Published, Side by Side
DeepSeek-V4-Pro (Preview): Benchmark Scores as Published
Every result the catalog holds for DeepSeek-V4-Pro (Preview): 69 tests in 13 areas, from 9 pages. 32 printed by DeepSeek itself, the rest by the company named beside each number. Each number is as printed; nothing is ranked or rescored.
Developer DeepSeekReleased 2026-04-24Context window 1MInput TextOutput TextWeights Open (MIT)Data version 2026103001, as of 2026-10-30
Knowledge Work
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| AutomationBench | 12.8% | DeepSeek, DeepSeek-V4-Flash-0731 model cardchecked 2026-10-15; printed on 1 more page | - |
| GDPval-AA v2 | 1307 | Thinking Machines, Thinking Machines Inkling model cardchecked 2026-10-15 | - |
| GDPval-AA (Elo) | 1554 | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: max |
| Vals Finance Agent v1.1 | 58.9% | NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15 | Tools: none |
| 62.3% | NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15 | Tools: web search | |
| GDPval rubrics (pointwise rubric score) | 70.32% | MiniMax, MiniMax M3checked 2026-10-15 | - |
| SpreadsheetBench v1 | 84.9% | MiniMax, MiniMax M3checked 2026-10-15 | Harness: Claude Code |
| GDPval (metric not stated) | 54.6% | NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15 | - |
| ProfBench (Search) | 59.9% | NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15 | Tools: web search |
Agentic Coding
Science and Math
Long Context
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| AA-LCR | 67.3% | NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15 | - |
| LongBench v2 | 62.1% | NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15 | - |
| CorpusQA 1M | 56.5% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: high; Context: 1M |
| 35.6% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: non-think; Context: 1M | |
| 62% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: max; Context: 1M | |
| MRCR 1M (MMR) | 44.7% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: non-think; Context: 1M |
| 83.5% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: max; Context: 1M | |
| 83.3% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: high; Context: 1M | |
| RULER at 1M | 94.2% | NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15 | Context: 1048576 |
Computer Use
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| OSWorld-Verified | 80.6% | MiniMax, MiniMax M3checked 2026-10-15 | - |
Cybersecurity
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| CyberGym | 52.7% | DeepSeek, DeepSeek-V4-Flash-0731 model cardchecked 2026-10-15; printed on 1 more page | - |
General Knowledge and Reasoning
Code Generation
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| LiveCodeBench v6 | 93.5% | Microsoft, Microsoft AI: Introducing MAI-Thinking-1checked 2026-10-19; printed on 1 more page | - |
| Codeforces rating | 3206 | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: max |
| 2919 | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: high | |
| LiveCodeBench (version not stated) | 89.8% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: high |
| 93.5% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: max | |
| 56.8% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: non-think | |
| IOI 2025 (total points) | 580.1 | NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15 | - |
Tool Use and Workflows
Web Search and Research
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| BrowseComp | 83.4% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15; printed on 3 more pages | Effort: max; Tools: enabled; exact tools in source |
| 80.4% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: high; Tools: enabled; exact tools in source |
Instruction Following
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| IFBench | 79.1 | NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15; printed on 1 more page | - |
| MultiChallenge | 64.1% | NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15 | - |
Multilingual
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| MMLU-ProX | 85.6% | NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15 | - |
| WMT24++ (XCOMET-XXL) | 85.9 | NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15 | - |
| Global-MMLU-Lite | 89.3% | Thinking Machines, Thinking Machines Inkling model cardchecked 2026-10-15 | - |
Safety and Alignment
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| FORTRESS (Adversarial) | 36% | Thinking Machines, Thinking Machines Inkling model cardchecked 2026-10-15 | - |
| FORTRESS (Benign) | 98.5% | Thinking Machines, Thinking Machines Inkling model cardchecked 2026-10-15 | - |
| StrongREJECT | 98.6% | Thinking Machines, Thinking Machines Inkling model cardchecked 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 |
|---|---|---|---|
| Cost per taskVals AI, USD, lower is better | $2.21 | 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-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-Pro-0813
- 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.