DeepSeek V4.1 Flash 1017 (max): benchmark scores, price and ranking
DeepSeek V4.1 Flash 1017 (max) from DeepSeek ranks #20 of 827 LLMs on the LLMs Tiger aggregate benchmark score (55.5), based on 13 benchmark results captured up to 2026-11-18.
- Organisation
- DeepSeek
- Type
- Open-weight / local model
- Open weights
- Yes
- Released
- 2026-10-17 (source: epoch)
- Cheapest API price
- $0.036 in / $0.500 out per million tokens (via openrouter, observed 2026-11-18T15:47Z)
- Aggregate rank
- #20 of 827 (score 55.5)
Relative to other models, DeepSeek V4.1 Flash 1017 (max) is strongest on GPQA Diamond (#22 of 302) and weakest on SimpleQA Verified (#40 of 83).
Across the 13 benchmarks where at least ten models are scored, DeepSeek V4.1 Flash 1017 (max) is in the top 10% on 2 and in the top half on 13.
Its cheapest listed input price, $0.036 per million tokens, is lower than that of 96% of the 443 models with a listed price.
DeepSeek V4.1 Flash 1017 (max) benchmark results
| Benchmark | Score | Rank | Source | Captured |
|---|---|---|---|---|
| GPQA Diamond | 93.2% | #22 of 302 | Epoch AI Benchmarking Hub | 2026-11-18 |
| OTIS Mock AIME | 99.4% | #17 of 206 | Epoch AI Benchmarking Hub | 2026-11-18 |
| APEX-Agents | 69.8% | #3 of 23 | Epoch AI Benchmarking Hub | 2026-11-18 |
| CritPt | 24.0% | #20 of 138 | Epoch AI Benchmarking Hub | 2026-11-18 |
| ALE-Bench | 1,622 | #15 of 93 | Epoch AI Benchmarking Hub | 2026-11-18 |
| ARC-AGI-2 | 78.1% | #20 of 97 | ARC Prize official leaderboard | 2026-11-18 |
| SciCode | 55.9% | #24 of 113 | Epoch AI Benchmarking Hub | 2026-11-18 |
| Mystery Games | 49.0% | #17 of 69 | Epoch AI Benchmarking Hub | 2026-11-18 |
| Chess Puzzles | 46.0% | #22 of 84 | Epoch AI Benchmarking Hub | 2026-11-18 |
| FrontierMath T1-3 | 71.2% | #27 of 78 | Epoch AI Benchmarking Hub | 2026-11-18 |
| FrontierMath T4 | 46.3% | #24 of 60 | Epoch AI Benchmarking Hub | 2026-11-18 |
| CursorBench | 47.7% | #10 of 22 | Epoch AI Benchmarking Hub | 2026-11-18 |
| SimpleQA Verified | 42.5% | #40 of 83 | Epoch AI Benchmarking Hub | 2026-11-18 |
About DeepSeek V4.1 Flash 1017 (max)
DeepSeek V4.1 Flash 1017 (max) is an open-weight model that can run on your own hardware from DeepSeek. It was released on 2026-10-17, according to Epoch AI. The cheapest API listing we track charges $0.036 per million input tokens and $0.500 per million output tokens, through openrouter. We hold 13 benchmark results for it across 4 areas, and 2 are in the top tenth of their boards.
Where DeepSeek V4.1 Flash 1017 (max) stands
On our aggregate score, DeepSeek V4.1 Flash 1017 (max) is #20 of 827 ranked models, in the top fifty. Its score is 55.5; the scale is centred on 50, so it sits 5.5 points above the typical model. That beats 98% of the ranked field. Among the 34 ranked DeepSeek models it is #1. The next one down is DeepSeek V4 Pro 0813 (max) at 52.9.
DeepSeek V4.1 Flash 1017 (max), benchmark by benchmark
On GPQA Diamond, DeepSeek V4.1 Flash 1017 (max) scores 93.2% and ranks #22 of 302. Only 21 models score higher, led by GPT-6.2 Astra (max) at 96.8%, which is 3.7 points ahead. It covers expert-level science questions. The board median is 75.1%, so DeepSeek V4.1 Flash 1017 (max) is 18.1 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-11-18.
On OTIS Mock AIME, DeepSeek V4.1 Flash 1017 (max) scores 99.4% and ranks #17 of 206. Only 16 models score higher, led by Claude Fable 5.1 (max) at 100.0%, which is 0.6 points ahead. It covers competition maths (AIME 2024-25 style). The board median is 66.4%, so DeepSeek V4.1 Flash 1017 (max) is 33.1 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-11-18.
DeepSeek V4.1 Flash 1017 (max) manages 69.8% on APEX-Agents, good for #3 of 23. Claude Sonnet 5.5 (max) leads at 75.5%, 5.7 points further up, and 87% of models sit below. The results either side are Claude Opus 5.5 (max) (73.5%) and GPT-6.1 Terra (max) (68.3%). It covers long professional agent tasks. The board median is 51.6%, so DeepSeek V4.1 Flash 1017 (max) is 18.2 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-11-18.
CritPt puts DeepSeek V4.1 Flash 1017 (max) at 24.0%, #20 of 138. That beats 86% of the models tested, but 12.0 points short of Gemini 4 Carbon (high) (36.0%). The results either side are Muse Spark 1.3 (max) (24.9%) and GPT-5.4 (xhigh) (23.4%). It covers unpublished research-level physics problems. The board median is 2.0%, so DeepSeek V4.1 Flash 1017 (max) is 22.0 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-11-18.
On ALE-Bench, DeepSeek V4.1 Flash 1017 (max) reaches 1,622. It ranks #15 of 93, ahead of 84% of the field, and 1,329 points behind GPT-6 Astra (max) at 2,951. The results either side are GPT-5.3 Codex (xhigh) (1,655) and Muse Spark Open (xhigh) (1,590). It covers AtCoder heuristic-optimisation contests (performance rating). The board median is 800, so DeepSeek V4.1 Flash 1017 (max) is 822 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-11-18.
DeepSeek V4.1 Flash 1017 (max) manages 78.1% on ARC-AGI-2, good for #20 of 97. GPT-6.2 Astra (max) leads at 95.8%, 17.8 points further up, and 79% of models sit below. The results either side are Inkling 2 Small (XHigh) (81.7%) and Dots3-Note Preview (Max) (76.8%). It covers abstract reasoning, ARC Prize public eval set. The board median is 40.1%, so DeepSeek V4.1 Flash 1017 (max) is 37.9 points above it. Source: ARC Prize official leaderboard, captured 2026-11-18.
FrontierMath T1-3 puts DeepSeek V4.1 Flash 1017 (max) at 71.2%, #27 of 78. That beats 65% of the models tested, but 23.9 points short of Claude Fable 5.2 (max) (95.1%). The results either side are Gemini 3.7 Flash (high) (71.6%) and Claude Opus 4.7 (max) (70.2%). It covers research-level maths, tiers 1-3. The board median is 61.6%, so DeepSeek V4.1 Flash 1017 (max) is 9.6 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-11-18.
On FrontierMath T4, DeepSeek V4.1 Flash 1017 (max) reaches 46.3%. It ranks #24 of 60, ahead of 60% of the field, and 53.7 points behind Claude Fable 5.2 (max) at 100.0%. The results either side are Claude Haiku 5.5 (max) (46.3%) and Muse Spark 1.3 (max) (46.3%). It covers research-level maths, tier 4. The board median is 32.9%, so DeepSeek V4.1 Flash 1017 (max) is 13.4 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-11-18.
DeepSeek V4.1 Flash 1017 (max) manages 47.7% on CursorBench, good for #10 of 22. Gemini 4 Carbon (high) leads at 60.4%, 12.7 points further up, and 55% of models sit below. The results either side are Claude Haiku 5.5 (max) (48.4%) and Claude Opus 5 (max) (46.6%). It covers real-world coding-agent tasks. The board median is 46.5%, so DeepSeek V4.1 Flash 1017 (max) is 1.2 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-11-18.
SimpleQA Verified puts DeepSeek V4.1 Flash 1017 (max) at 42.5%, #40 of 83. That beats 52% of the models tested, but 51.5 points short of Claude Fable 5.2 (max) (94.0%). The results either side are deepseek-v4.1-flash-max (42.9%) and Doubao Seed Evolving (Nov 2026, high) (41.9%). It covers short factual questions. The board median is 41.4%, so DeepSeek V4.1 Flash 1017 (max) is 1.1 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-11-18.
The full table above also lists the other 3 results, including boards with fewer than ten models, where a rank says little.
Strengths and weaknesses
Software engineering: solid, ahead of 72% of the field on average across 3 benchmarks. Its best board here is ALE-Bench (#15 of 93) and its weakest is CursorBench (#10 of 22). Reasoning, maths and science: strong, ahead of 79% of the field on average across 6 benchmarks. Its best board here is GPQA Diamond (#22 of 302) and its weakest is FrontierMath T4 (#24 of 60). General knowledge and chat: solid, ahead of 67% of the field on average across 3 benchmarks. Its best board here is Mystery Games (#17 of 69) and its weakest is SimpleQA Verified (#40 of 83).
Price and value
The median input price across the 443 priced models is $0.482 per million tokens, so DeepSeek V4.1 Flash 1017 (max) costs less than the typical listing. We found no cheaper model with a similar or better aggregate score, so DeepSeek V4.1 Flash 1017 (max) is the lowest-priced option at its level. Similar scores cost more on Claude Haiku 5.5 (max) ($0.100) and Kimi K3 (max) ($0.192).
Release history
The DeepSeek model released before it was deepseek-v4.1-flash-max on 2026-09-09, which scores 51.7.
Reading these numbers
The 13 scores on this page come from 2 sources: Epoch AI Benchmarking Hub (12), ARC Prize official leaderboard (1). 1 of them come from an official leaderboard and 12 from datasets or third-party runs. The leaderboard also lists 5 other reasoning-effort settings of this model, which redirect here. This page shows the highest.
DeepSeek V4.1 Flash 1017 (max) is also listed at 5 other reasoning-effort settings; this page shows the highest-effort row, which is the one the leaderboard keeps by default.
Models ranked nearby
- #17 Muse Wonder (max) (56.1)
- #18 GPT-5.5 (xhigh) (56.1)
- #19 GPT-5.4 (xhigh) (55.9)
- #21 Kimi K3.1 Pro (max) (55.5)
- #22 gemini-3.1-pro-preview-high (55.3)
- #23 Kimi K3.1 Pro (Together, max) (55.3)
More from DeepSeek
- DeepSeek V4 Pro 0813 (max) (#52, 52.9)
- DeepSeek-V3.2-Exp (high) (#77, 52.4)
- DeepSeek V4 Flash 0731 (max) (#82, 52.3)
- DeepSeek-V2.5-1210 (#125, 51.7)
- deepseek-v4.1-flash-max (#131, 51.7)
- DeepSeek Chat V2 0628 (#156, 51.4)
Compare it on the live leaderboard → · How the aggregate score works