DeepSeek V4.1 Flash
DeepSeek · released
DeepSeek's newest model: a sparse 552B-parameter design that activates only a sliver of itself per token, so it is cheap and fast while offering a 1M-token context and image input. On agentic and coding results it outperforms the lab's larger V4 Pro.
Best for: Cheap agentic work · MIT-licensed open weights · High-throughput coding
- Intelligence Index
- 39.5
- #19 of 38 · at Max effort
- Price per 1M tokens
- $0.30 in $1.2 out
- Blended $0.52 at 3:1
- Context window
- 1M
- Up to 384K output
- Output speed
- 216 tokens/s
- $0.27 measured cost per task
Capability profile
One 0–100 score per area, placing this model between the weakest and strongest tracked model across that area's benchmarks.
- Overall64 · #18 of 38
Based on 3 of 3 benchmarks
- Coding59 · #14 of 39
Based on 5 of 5 benchmarks
- Agents & knowledge work64 · #17 of 40
Based on 6 of 6 benchmarks
- Reasoning & science51 · #21 of 39
Based on 2 of 2 benchmarks
- Knowledge & reliability62 · #22 of 39
Based on 2 of 2 benchmarks
- Vision65 · #17 of 31
Based on 1 of 2 benchmarks
Reasoning effort trade-off
The same model at each effort setting. Lower settings give up a few points for a fraction of the cost and wait.
| Setting | Intelligence Index | Cost / task | First answer |
|---|---|---|---|
| Max | 39.5 | $0.27 | 10s |
| No reasoning | 24.7 | $0.15 | 1.1s |
Measured by Artificial Analysis. First answer is the median wait until the first answer token, including thinking time.
Benchmark results
All third-party results on record. Rank is among the tracked models that have been tested on that benchmark.
Overall
#19 of 38· Max effort· Artificial Analysis
- Vals Index51.3%
#15 of 26· High effort· $0.33 per test· Vals AI
- Arena · Text1474 ±7
#13 of 33· Max effort· 8,728 votes· Arena
Coding
- Terminal-Bench 4.026.8%
#15 of 38· Max effort· Artificial Analysis
- SciCode51.9%
#20 of 38· Max effort· Artificial Analysis
- Arena · WebDev1619 ±11
#16 of 34· Max effort· 3,547 votes· Arena
- Vibe Code Bench84.7%
#12 of 26· High effort· Vals AI
- Code Migration46.9%
#9 of 26· High effort· Vals AI
Agents & knowledge work
- GDPval-AA55.0
#14 of 39· Max effort· Artificial Analysis
- Arena · Agent+4.0 ±0.5
#8 of 29· Max effort· Arena
- Finance Agent v253.5%
#15 of 26· High effort· Vals AI
#18 of 26· High effort· Vals AI
- Legal Research Bench41.4%
#13 of 26· High effort· Vals AI
- Tax Agent Bench62.5%
#16 of 26· High effort· Vals AI
Reasoning & science
- Humanity's Last Exam39.2%
#20 of 39· Max effort· Artificial Analysis
- CritPt14.3%
#21 of 39· Max effort· Artificial Analysis
- ARC-AGI-2saturating72.9%
#8 of 16· Max effort· $0.13 per task· ARC Prize
- ARC-AGI-3Not tested
- GPQA DiamondsaturatingNot tested
Knowledge & reliability
#26 of 39· Max effort· Artificial Analysis
- AA-LCR84.0%
#6 of 39· Max effort· Artificial Analysis
Vision
- MMMU-Pro77.0%
#15 of 27· Max effort· Artificial Analysis
- Arena · VisionNot tested
Reported by DeepSeek
Numbers the lab published at launch, run on its own harness and settings. Useful context, but not independently verified and never used in the rankings on this site.
- DeepSWE v1.1 · mini-SWE harness74.2%
- AutomationBench54.8%
- Terminal-Bench 4.0 · own harness31.2%
- GPQA Diamond90.9%
Specifications
- Lab
- DeepSeek
- Released
- Sep 10, 2026
- Availability
- Generally available
- Context window
- 1,000,000 tokens
- Max output
- 384,000 tokens
- Accepts
- Text, Images
- Knowledge cutoff
- Not published
- Weights
- Open
- Licence
- MIT
- Parameters
- 552B total · 8–16B active
- Reasoning control
- Thinking on by default · can be switched off
- API model ID
deepseek-flash
Pricing
- Input / 1M tokens
- $0.30
- Cached input / 1M
- <$0.01
- Output / 1M tokens
- $1.2
- Blended (3:1)
- $0.52
- Measured cost / task
- $0.27
Peak-hour list price. Off-peak requests cost half.
Official sources
Closest alternatives
The models nearest to this one on overall score.
23 benchmarks tracked. See methodology and sources.