DeepSeek V4.1 Flash vs DeepSeek V4.1 Flash 1017
Vals Science Index?
DeepSeek V4.1 Flash29.7
DeepSeek V4.1 Flash 1017—
| Benchmark | DeepSeek V4.1 Flash | DeepSeek V4.1 Flash 1017 |
|---|---|---|
| Humanity's Last Exam | 10.8 | 43.3 |
| Arena Text · overall | 1,475 · max | 1,436 |
| Arena Text · Chinese | 1,503.3 · max | 1,462.8 |
| Arena Text · English | 1,473.1 · max | 1,440.8 |
| Arena Text · French | 1,463.1 · max | 1,439 |
| Arena Text · German | 1,476 · max | 1,418.6 |
| Arena Text · instruction following | 1,473.4 · max | 1,420.8 |
| Arena Text · Japanese | 1,420.6 · max | 1,388.9 |
| Arena Text · Korean | 1,440.6 · max | 1,382.3 |
| Arena Text · non-English | 1,448.7 · max | 1,419.2 |
| Arena Text · Polish | 1,453.4 · max | 1,435.3 |
| Arena Text · Russian | 1,465.4 · max | 1,427 |
| Arena Text · Spanish | 1,458.6 · max | 1,435.5 |
| AA Omniscience Index | -10.2 | 4.7 |
| CritPt (physics) | 0.3 | 24 |
| AA Intelligence Index 4.3.2 | 24.7 | 45.1 |
| AA Omniscience · Accuracy | 28.5 | 42.9 |
| AA Omniscience · Hallucination Rate | 54 | 66.8 |
| AA-LCR (long context) 1.1 | 55.3 | 82.7 |
| GDP.pdf · AA all-pass | 3.8 | 17.2 |
| ARC-AGI-1 · public evaluation | 98 · max | 19.3 · none |
| ARC-AGI-1 · semi-private | 94.5 · max | 11.8 · none |
| ARC-AGI-2 · public evaluation | 81.7 · max | 0 · none |
| ARC-AGI-2 · semi-private | 72.9 · max | 2.1 · none |
| MathArena · ArXivMath · April 2026 | 45 · max | 47.5 · max |
| MathArena · ArXivMath · June 2026 | 57.6 · max | 43.8 |
| MathArena · ArXivMath · May 2026 | 58.3 · max | 55.8 |
| MathArena · BrokenArXiv · June 2026 | 39.4 · max | 16.7 |
| MathArena · BrokenArXiv · May 2026 | 28.5 · max | 19.5 |
| Vals Legal Research Bench | 41.4 | 45.2 |
| SimpleQA Verified v1 | 42.9 · max | — |
| MLCR-AA | 22.8 · max | — |
| ARC-AGI-3 · semi-private | 0.9 · max | — |
| EQ-Bench 3 | — | 1,491.3 |
| Chess Puzzles | 48 · max | — |
| FrontierMath Tier 4 (v2) | 26.8 · max | — |
| FrontierMath Tiers 1–3 (v2) | 67.4 · max | — |
| Mystery Game Puzzles | 43 · max | — |
| MathArena · Apex | — | 27.1 · max |
| MathArena · Apex Shortlist | — | 89.4 · max |
| MathArena · ArXivMath · August 2026 | 43.9 · max | — |
| MathArena · ArXivMath · February 2026 | — | 43.8 · max |
| MathArena · ArXivMath · March 2026 | — | 52.5 · max |
| MathArena · ArXivMath · September 2026 | 40.6 · max | — |
| MathArena · BrokenArXiv · April 2026 | — | 23.8 · max |
| MathArena · BrokenArXiv · August 2026 | 54.8 · max | — |
| MathArena · BrokenArXiv · February 2026 | — | 17.7 · max |
| MathArena · BrokenArXiv · March 2026 | — | 19.6 · max |
| MathArena · BrokenArXiv · September 2026 | 54.7 · max | — |
| MathArena · ArXivLean · September 2026 | 5.7 · max | — |
| Vals LegalBench | 83.3 | — |
| Vals MedCode | 41.2 | — |
| Vals MedScribe | 85.5 | — |
| Vals Science Index | 29.7 | — |
Specs
| Spec | DeepSeek V4.1 Flash | DeepSeek V4.1 Flash 1017 |
|---|---|---|
| Input, per 1M | !$0.30 | $0.30 |
| Output, per 1M | !$1.20 | $1.20 |
| Cost per task? | $0.22 | $0.26 |
| Speed, tok/s | 201 | 167 |
| Context | 1M | 1M |
| Weights | Open | Open |
| Released | Sep 2026 | Oct 2026 |
| Memory (4-bit) | — | 458 GB |
| Licence | — | MIT |
| Commercial use | — | Allowed |
| See images | ? | ✓ |
| Read PDFs | ? | — |
| Hear audio | ? | — |
| Watch video | ? | — |
| Call tools | ✓ | ✓ |
| Force a tool call | ? | ✓ |
| Return strict JSON | ✓ | ✓ |
| Search the web | ? | ? |
| Run code | ? | ? |
| Use a computer | ? | ? |
What breaks when you switch
DeepSeek V4.1 Flash → DeepSeek V4.1 Flash 1017
Pass back reasoning_content · Chat Completions›
With thinking and tools, return all prior reasoning_content even on turns without tool calls or the API returns HTTP 400.
Disable thinking mode first · Chat Completions›
Required and named tool choices return HTTP 400 in thinking mode; disable thinking first.
Parameters have no effect›
In thinking mode, temperature, presence_penalty and frequency_penalty have no effect.
Values below 0.95›
Flash clamps top_p below 0.95 to 0.95 while thinking and ignores it in non-thinking mode, using 1.0.
Stateless API · Responses›
Responses does not support previous_response_id or conversation; preserve conversation history in your client.
DeepSeek V4.1 Flash 1017 → DeepSeek V4.1 Flash