DeepSeek-V4-Flash vs Devstral 2
Wins 5 of 6 areas
Coding · Agents · Reasoning · Long documents · Following instructions
Wins 0 of 6 areas
—
DeepSeek-V4-Flash is the stronger all-rounder.
Scores updated · 22 tests both models report · How we compare
Where each one wins
Tests won in each of the six 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 software60DeepSeek-V4-Flash6 of 6 tests
- AgentsCarrying out multi-step tasks on its own30DeepSeek-V4-Flash3 of 3 tests
- ReasoningHard problems that need careful thinking30DeepSeek-V4-Flash3 of 3 tests
- Long documentsFinding answers in very long texts10DeepSeek-V4-Flash1 of 1 test
- Following instructionsDoing exactly what it is asked10DeepSeek-V4-Flash1 of 1 test
- FactsGetting facts right instead of making them up11Even1 each
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.
DeepSeek-V4-Flash costs 27% less for the same work.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where DeepSeek-V4-Flash pulls ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+48.4points ahead
- Reasons across sets of long documentsAA-LCR+47.4points ahead
- Real work tasks from 44 professionsGDPVal+44.5points ahead
Where Devstral 2 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+6.4points ahead
Every test, side by side
All 22 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingDeepSeek-V4-Flash
- Terminal-Bench 2.1DeepSeek-V4-Flash by 48.478.730.3+48.4
- SWE-bench VerifiedDeepSeek-V4-Flash by 25.27953.8+25.2
- LMArena · WebDevDeepSeek-V4-Flash by 236 rating points14301194+236 rating
- SciCodeDeepSeek-V4-Flash by 17.550.332.8+17.5
- Terminal-Bench HardDeepSeek-V4-Flash by 16.735.618.9+16.7
- SWE-bench MultilingualDeepSeek-V4-Flash by 1273.361.3+12
AgentsDeepSeek-V4-Flash
- GDPValDeepSeek-V4-Flash by 44.546.92.4+44.5
- τ-Bench V3 · BankingDeepSeek-V4-Flash by 28.939.410.5+28.9
- Terminal-Bench 4.0DeepSeek-V4-Flash by 12.112.10+12.1
ReasoningDeepSeek-V4-Flash
- Humanity's Last ExamDeepSeek-V4-Flash by 3538.63.6+35
- GPQA DiamondDeepSeek-V4-Flash by 31.490.859.4+31.4
- CritPtDeepSeek-V4-Flash by 16.616.60+16.6
FactsEven
- AA-Omniscience · AccuracyDeepSeek-V4-Flash by 19.640.420.8+19.6
- AA-Omniscience · Non-hallucinationDevstral 2 by 6.48.314.7+6.4
Long documentsDeepSeek-V4-Flash
- AA-LCRDeepSeek-V4-Flash by 47.479.732.3+47.4
Following instructionsDeepSeek-V4-Flash
- IFBenchDeepSeek-V4-Flash by 41.179.238.1+41.1
Other results6 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.
- τ²-Bench Telecom (AA run)DeepSeek-V4-Flash by 70.19524.9+70.1
- Artificial Analysis Coding IndexDeepSeek-V4-Flash by 37.869.131.3+37.8
- AA Agentic IndexDeepSeek-V4-Flash by 36.841.74.9+36.8
- AA-OmniscienceDeepSeek-V4-Flash by 32.4-14.3-46.7+32.4
- AA IntelligenceDeepSeek-V4-Flash by 25.734.38.6+25.7
- Terminal-Bench 2.0DeepSeek-V4-Flash by 24.356.932.6+24.3
Questions people ask
Which is better, DeepSeek-V4-Flash or Devstral 2?
DeepSeek-V4-Flash wins five of the six areas where both have results: coding, agents, reasoning, long documents and following instructions. Devstral 2 wins none. They are level on facts.
Which is better for coding?
DeepSeek-V4-Flash. It wins 6 of the 6 coding tests both models report; Devstral 2 wins none.
Which is cheaper?
DeepSeek-V4-Flash costs $0.44 per million input tokens and $1.32 per million output tokens; Devstral 2 costs $0.40 and $2.00. That makes DeepSeek-V4-Flash about 27% cheaper for the same work.
How do you compare the two?
We use the 22 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 6 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.