DeepSeek-V4-Flash vs Qwen3.8 Flash Next
Wins 1 of 7 areas
Math
Wins 3 of 7 areas
Coding · Agents · Following instructions
Qwen3.8 Flash Next wins more areas, narrowly.DeepSeek-V4-Flash is better at math.
Scores updated · 49 tests both models report · How we compare
Where each one wins
Tests won in each of the seven areas where both have results. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.
- CodingWriting and fixing software16Qwen3.8 Flash Next6 of 9 tests · 2 ties
- AgentsCarrying out multi-step tasks on its own04Qwen3.8 Flash Next4 of 4 tests
- Following instructionsDoing exactly what it is asked02Qwen3.8 Flash Next2 of 2 tests
- MathCompetition and research-level math10DeepSeek-V4-Flash1 of 1 test
- ReasoningHard problems that need careful thinking11Even1 each · 2 ties
- FactsGetting facts right instead of making them up11Even1 each
- Long documentsFinding answers in very long texts00Even0 each · 1 tie
Math and long documents 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.
Qwen3.8 Flash Next costs 65% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Qwen3.8 Flash Next pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+46.4points ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+15.8points ahead
- Fixes real GitHub issues, working as an agentLiveBench · Agentic Coding+14.8points ahead
Where DeepSeek-V4-Flash pulls ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+15.9points ahead
- Research-level physics problemsCritPt+5.5points ahead
- Programming problems, refreshed regularlyLiveBench · Coding+2.4points ahead
Every test, side by side
All 49 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingQwen3.8 Flash Next
- LMArena · WebDevQwen3.8 Flash Next by 199 rating points14311630+199 rating
- LiveBench · Agentic CodingQwen3.8 Flash Next by 14.846.861.6+14.8
- SWE-bench ProQwen3.8 Flash Next by 9.952.662.5+9.9
- SWE-bench MultilingualQwen3.8 Flash Next by 7.773.381+7.7
- Terminal-Bench 2.1Qwen3.8 Flash Next by 7.478.786.1+7.4
- LiveBench · CodingDeepSeek-V4-Flash by 2.47572.5+2.4
- LiveCodeBench v6Qwen3.8 Flash Next by 1.390.691.9+1.3
- SciCodetie50.350.6tie
- SWE-bench Verifiedtie7978.7tie
AgentsQwen3.8 Flash Next
- BrowseCompQwen3.8 Flash Next by 15.873.289+15.8
- Terminal-Bench 4.0Qwen3.8 Flash Next by 13.212.125.3+13.2
- GDPValQwen3.8 Flash Next by 9.746.756.4+9.7
- τ-Bench V3 · BankingQwen3.8 Flash Next by 639.445.4+6
ReasoningEven
- CritPtDeepSeek-V4-Flash by 5.516.611.1+5.5
- GPQA DiamondQwen3.8 Flash Next by 1.590.892.3+1.5
- LiveBench · Reasoningtie86.687.4tie
- Humanity's Last Examtie38.638tie
FactsEven
- AA-Omniscience · Non-hallucinationQwen3.8 Flash Next by 46.48.354.7+46.4
- AA-Omniscience · AccuracyDeepSeek-V4-Flash by 15.940.424.5+15.9
Following instructionsQwen3.8 Flash Next
- LiveBench · Instruction FollowingQwen3.8 Flash Next by 11.665.577.1+11.6
- IFBenchQwen3.8 Flash Next by 2.179.281.3+2.1
Other results26 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.
- GDPval-AA v2Qwen3.8 Flash Next by 372 rating points11891561+372 rating
- Agents' Last ExamQwen3.8 Flash Next by 2625.251.2+26
- τ³-Bench BankingQwen3.8 Flash Next by 21.622.944.5+21.6
- Terminal Bench 4 [offline]Qwen3.8 Flash Next by 17.23.520.7+17.2
- AA Agentic IndexQwen3.8 Flash Next by 14.741.756.4+14.7
- LMArena v3 proxyDeepSeek-V4-Flash by 14.649.835.2+14.6
- JobBenchQwen3.8 Flash Next by 14.441.355.7+14.4
- DeepSWE [offline]Qwen3.8 Flash Next by 13.342.355.6+13.3
- tau2-bench telecomDeepSeek-V4-Flash by 7.898.290.4+7.8
- NL2RepoDeepSeek-V4-Flash by 6.154.248.1+6.1
- BrowseComp [offline]Qwen3.8 Flash Next by 68389+6
- AA IntelligenceQwen3.8 Flash Next by 5.534.339.8+5.5
- livebench_data_analysisDeepSeek-V4-Flash by 5.179.374.2+5.1
- AA-OmniscienceQwen3.8 Flash Next by 4.6-14.3-9.7+4.6
- livebench_languageDeepSeek-V4-Flash by 4.679.274.6+4.6
- DeepSWE 1.1Qwen3.8 Flash Next by 4.354.458.7+4.3
- Artificial Analysis Coding IndexQwen3.8 Flash Next by 469.173.1+4
- HLE, text-onlyQwen3.8 Flash Next by 3.835.839.6+3.8
- ToolathlonQwen3.8 Flash Next by 3.270.373.5+3.2
- Toolathlon VerifiedQwen3.8 Flash Next by 3.270.373.5+3.2
- SciCode, subtaskDeepSeek-V4-Flash by 2.749.346.6+2.7
- tau2-bench airlineDeepSeek-V4-Flash by 2.783.580.8+2.7
- MMLU-ProXDeepSeek-V4-Flash by 2.585.583+2.5
- tau2-bench retailDeepSeek-V4-Flash by 2.273.871.6+2.2
- SciCode (subtask)DeepSeek-V4-Flash by 1.648.246.6+1.6
- WMT24++tie85.586.2tie
Questions people ask
Which is better, DeepSeek-V4-Flash or Qwen3.8 Flash Next?
Qwen3.8 Flash Next wins three of the seven areas where both have results: coding, agents and following instructions. DeepSeek-V4-Flash wins math. They are level on reasoning, facts and long documents.
Which is better for coding?
Qwen3.8 Flash Next. It wins 6 of the 9 coding tests both models report; DeepSeek-V4-Flash wins 1, and 2 are ties.
Which is cheaper?
DeepSeek-V4-Flash costs $0.44 per million input tokens and $1.32 per million output tokens; Qwen3.8 Flash Next costs $0.15 and $0.47. That makes Qwen3.8 Flash Next about 65% cheaper for the same work.
How do you compare the two?
We use the 49 benchmark tests both models have published scores on. The verdict counts the 23 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 26 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.