DeepSeek-V4.1-Flash vs Qwen3.8 27B
Wins 5 of 8 areas
Coding · Agents · Reasoning · Math · Long documents
Wins 1 of 8 areas
Following instructions
DeepSeek-V4.1-Flash is the stronger all-rounder.Qwen3.8 27B is better at following instructions.
Scores updated · 41 tests both models report · How we compare
Where each one wins
Tests won in each of the eight areas we test. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.
- ReasoningHard problems that need careful thinking50DeepSeek-V4.1-Flash5 of 6 tests · 1 tie
- CodingWriting and fixing software50DeepSeek-V4.1-Flash5 of 5 tests
- AgentsCarrying out multi-step tasks on its own20DeepSeek-V4.1-Flash2 of 2 tests
- Long documentsFinding answers in very long texts10DeepSeek-V4.1-Flash1 of 2 tests · 1 tie
- MathCompetition and research-level math10DeepSeek-V4.1-Flash1 of 1 test
- Following instructionsDoing exactly what it is asked01Qwen3.8 27B1 of 1 test
- FactsGetting facts right instead of making them up11Even1 each
- Images and chartsUnderstanding pictures, charts and video11Even1 each · 1 tie
Math and following instructions rest on a single test each.
What it costs
Prices per million tokens, roughly 750,000 words. The bars show the cost of a million tokens read plus a million written.
DeepSeek-V4.1-Flash costs 57% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where DeepSeek-V4.1-Flash pulls ahead
- Abstract visual puzzles that people can solveARC-AGI-2+35.7points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+30.8points ahead
- Complex command-line tasks across many fieldsTerminal-Bench 4.0+28.2points ahead
Where Qwen3.8 27B pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+66.2points ahead
- Reasoning about charts from research papersCharXiv (RQ)+21.2points ahead
- Follows detailed instructions exactlyLiveBench · Instruction Following+2.7points ahead
Every test, side by side
All 41 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingDeepSeek-V4.1-Flash
- LiveBench · Agentic CodingDeepSeek-V4.1-Flash by 15.977.361.4+15.9
- Terminal-Bench 2.1DeepSeek-V4.1-Flash by 10.890.679.8+10.8
- SciCodeDeepSeek-V4.1-Flash by 9.355.946.6+9.3
- LiveBench · CodingDeepSeek-V4.1-Flash by 4.38075.7+4.3
- LMArena · WebDevDeepSeek-V4.1-Flash by 31 rating points16211590+31 rating
AgentsDeepSeek-V4.1-Flash
- Terminal-Bench 4.0DeepSeek-V4.1-Flash by 28.233.85.6+28.2
- GDPValDeepSeek-V4.1-Flash by 95546+9
ReasoningDeepSeek-V4.1-Flash
- ARC-AGI-2DeepSeek-V4.1-Flash by 35.778.142.4+35.7
- CritPtDeepSeek-V4.1-Flash by 18.6245.4+18.6
- Humanity's Last ExamDeepSeek-V4.1-Flash by 9.443.333.9+9.4
- LiveBench · ReasoningDeepSeek-V4.1-Flash by 6.786.780+6.7
- SimpleBenchDeepSeek-V4.1-Flash by 6.566.760.2+6.5
- GPQA Diamondtie90.990.5tie
FactsEven
- AA-Omniscience · Non-hallucinationQwen3.8 27B by 66.23.569.7+66.2
- AA-Omniscience · AccuracyDeepSeek-V4.1-Flash by 30.846.415.6+30.8
Images and chartsEven
- CharXiv (RQ)Qwen3.8 27B by 21.26990.2+21.2
- MMMU-ProDeepSeek-V4.1-Flash by 4.380.676.3+4.3
- LMArena · Visiontie12781271tie
MathDeepSeek-V4.1-Flash
- LiveBench · MathematicsDeepSeek-V4.1-Flash by 7.193.386.2+7.1
Long documentsDeepSeek-V4.1-Flash
- LongBench v2DeepSeek-V4.1-Flash by 16.76952.3+16.7
- AA-LCRtie82.782tie
Following instructionsQwen3.8 27B
- LiveBench · Instruction FollowingQwen3.8 27B by 2.77072.7+2.7
Other results19 tests, not counted
Tests outside the eight areas. They are not counted above: several are summary scores built from other tests, or the same test under another name.
- CritPt v2 FullDeepSeek-V4.1-Flash by 4050.610.6+40
- JobBenchDeepSeek-V4.1-Flash by 35.168.533.4+35.1
- DeepSWE 1.1DeepSeek-V4.1-Flash by 3274.242.2+32
- GDPval-AA 2.1DeepSeek-V4.1-Flash by 319 rating points17461427+319 rating
- Terminal-Bench 4.0DeepSeek-V4.1-Flash by 25.631.25.6+25.6
- NL2RepoDeepSeek-V4.1-Flash by 21.76442.3+21.7
- CritPt v2 HardDeepSeek-V4.1-Flash by 21254+21
- Terminal-Bench-Science 0.1DeepSeek-V4.1-Flash by 12.912.90+12.9
- Agents' Last ExamQwen3.8 27B by 11.131.842.9+11.1
- HLE-DiamondDeepSeek-V4.1-Flash by 1025.315.3+10
- SWEBench Pro v2 (Hard)DeepSeek-V4.1-Flash by 8.682.473.8+8.6
- livebench_languageDeepSeek-V4.1-Flash by 6.881.274.3+6.8
- ARC-AGI-1DeepSeek-V4.1-Flash by 693.587.5+6
- AA IntelligenceDeepSeek-V4.1-Flash by 5.839.533.7+5.8
- Chartography originalDeepSeek-V4.1-Flash by 5.717.912.2+5.7
- AA-OmniscienceDeepSeek-V4.1-Flash by 4.7-5.3-10+4.7
- BabyVisionDeepSeek-V4.1-Flash by 489.685.6+4
- livebench_data_analysisDeepSeek-V4.1-Flash by 2.779.376.6+2.7
- GDP (Surge AI)tie17.216.4tie
Questions people ask
Which is better, DeepSeek-V4.1-Flash or Qwen3.8 27B?
DeepSeek-V4.1-Flash wins five of the eight areas we test: coding, agents, reasoning, math and long documents. Qwen3.8 27B wins following instructions. They are level on facts and images and charts.
Which is better for coding?
DeepSeek-V4.1-Flash. It wins 5 of the 5 coding tests both models report; Qwen3.8 27B wins none.
Which is cheaper?
DeepSeek-V4.1-Flash costs $0.30 per million input tokens and $1.20 per million output tokens; Qwen3.8 27B costs $0.50 and $3.00. That makes DeepSeek-V4.1-Flash about 57% cheaper for the same work.
How do you compare the two?
We use the 41 benchmark tests both models have published scores on. The verdict counts the 22 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 19 are listed but not counted, because several are summary scores or repeat a test. Each score is the one shown on the model's own page.