DeepSeek-V3.2 vs Step 3.5 Flash
Wins 3 of 7 areas
Reasoning · Facts · Long documents
Wins 4 of 7 areas
Coding · Agents · Math · Following instructions
Step 3.5 Flash wins more areas, narrowly.DeepSeek-V3.2 is better at facts.
Scores updated · 33 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.
- MathCompetition and research-level math03Step 3.5 Flash3 of 3 tests
- CodingWriting and fixing software13Step 3.5 Flash3 of 4 tests
- AgentsCarrying out multi-step tasks on its own02Step 3.5 Flash2 of 2 tests
- Following instructionsDoing exactly what it is asked01Step 3.5 Flash1 of 1 test
- FactsGetting facts right instead of making them up20DeepSeek-V3.22 of 2 tests
- ReasoningHard problems that need careful thinking10DeepSeek-V3.21 of 3 tests · 2 ties
- Long documentsFinding answers in very long texts10DeepSeek-V3.21 of 1 test
Long documents 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.
Step 3.5 Flash costs 43% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Step 3.5 Flash pulls ahead
- Real work tasks from 44 professionsGDPVal+17.8points ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+17.6points ahead
- Olympiad math problems with short answersIMOAnswerBench+7.1points ahead
Where DeepSeek-V3.2 pulls ahead
- Reasons across sets of long documentsAA-LCR+23.3points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+9.4points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+8.3points ahead
Every test, side by side
All 33 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingStep 3.5 Flash
- Terminal-Bench HardDeepSeek-V3.2 by 8.335.627.3+8.3
- LiveCodeBench v6Step 3.5 Flash by 3.183.386.4+3.1
- SciCodeStep 3.5 Flash by 1.538.940.4+1.5
- SWE-bench VerifiedStep 3.5 Flash by 1.373.174.4+1.3
AgentsStep 3.5 Flash
- GDPValStep 3.5 Flash by 17.89.827.6+17.8
- BrowseCompStep 3.5 Flash by 17.651.469+17.6
ReasoningDeepSeek-V3.2
- Humanity's Last ExamDeepSeek-V3.2 by 3.524.621.1+3.5
- GPQA Diamondtie8483.1tie
- CritPttie2.92.5tie
FactsDeepSeek-V3.2
- AA-Omniscience · AccuracyDeepSeek-V3.2 by 9.43323.6+9.4
- AA-Omniscience · Non-hallucinationDeepSeek-V3.2 by 317.314.3+3
MathStep 3.5 Flash
- IMOAnswerBenchStep 3.5 Flash by 7.178.385.4+7.1
- AIME 2026Step 3.5 Flash by 2.594.296.7+2.5
- HMMT Feb. 2026Step 3.5 Flash by 2.384.186.4+2.3
Following instructionsStep 3.5 Flash
- IFBenchStep 3.5 Flash by 3.960.764.6+3.9
Other results17 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.
- AA Agentic IndexStep 3.5 Flash by 33.718.352+33.7
- AA-OmniscienceDeepSeek-V3.2 by 19.3-22.5-41.8+19.3
- Artificial Analysis Coding IndexDeepSeek-V3.2 by 12.644.231.6+12.6
- ResearchRubricsStep 3.5 Flash by 9.555.865.3+9.5
- GAIA (no file)Step 3.5 Flash by 9.475.184.5+9.4
- τ²-BenchStep 3.5 Flash by 7.980.388.2+7.9
- HMMT Feb. 2025Step 3.5 Flash by 5.992.598.4+5.9
- AA IntelligenceDeepSeek-V3.2 by 4.921.516.6+4.9
- Terminal-Bench 2.0Step 3.5 Flash by 4.646.451+4.6
- AIME 2025Step 3.5 Flash by 4.293.197.3+4.2
- HMMT Nov. 2025Step 3.5 Flash by 49094+4
- τ²-Bench Telecom (AA run)Step 3.5 Flash by 3.890.694.4+3.8
- BrowseComp-ZHStep 3.5 Flash by 1.96566.9+1.9
- browsecomp_with_context_managerStep 3.5 Flash by 1.467.669+1.4
- MMLU-Protie8584.4tie
- xbench-DeepSearchtie55.756.3tie
- xbench-DeepSearch (2025.10)tie55.756.3tie
Questions people ask
Which is better, DeepSeek-V3.2 or Step 3.5 Flash?
Step 3.5 Flash wins four of the seven areas where both have results: coding, agents, math and following instructions. DeepSeek-V3.2 wins reasoning, facts and long documents.
Which is better for coding?
Step 3.5 Flash. It wins 3 of the 4 coding tests both models report; DeepSeek-V3.2 wins 1.
Which is cheaper?
DeepSeek-V3.2 costs $0.28 per million input tokens and $0.42 per million output tokens; Step 3.5 Flash costs $0.10 and $0.30. That makes Step 3.5 Flash about 43% cheaper for the same work.
How do you compare the two?
We use the 33 benchmark tests both models have published scores on. The verdict counts the 16 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 17 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.