DeepSeek-V4-Pro vs Kimi K2.5
Wins 6 of 7 areas
Coding · Agents · Reasoning · Facts · Math · Following instructions
Wins 1 of 7 areas
Long documents
DeepSeek-V4-Pro is the stronger all-rounder.Kimi K2.5 is better at long documents.
Scores updated · 48 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 software80DeepSeek-V4-Pro8 of 8 tests
- AgentsCarrying out multi-step tasks on its own50DeepSeek-V4-Pro5 of 5 tests
- FactsGetting facts right instead of making them up31DeepSeek-V4-Pro3 of 4 tests
- ReasoningHard problems that need careful thinking20DeepSeek-V4-Pro2 of 3 tests · 1 tie
- MathCompetition and research-level math20DeepSeek-V4-Pro2 of 3 tests · 1 tie
- Following instructionsDoing exactly what it is asked10DeepSeek-V4-Pro1 of 1 test
- Long documentsFinding answers in very long texts01Kimi K2.51 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.
DeepSeek-V4-Pro costs 64% 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-Pro pulls ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+18.3points ahead
- Customer-service tasks in a simulated bankτ-Bench V3 · Banking+15.9points ahead
- Real work tasks from 44 professionsGDPVal+15.7points ahead
Where Kimi K2.5 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+28.4points ahead
- Reasons across sets of long documentsAA-LCR+3.3points ahead
Every test, side by side
All 48 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingDeepSeek-V4-Pro
- Terminal-Bench 2.1DeepSeek-V4-Pro by 18.36445.7+18.3
- Terminal-Bench HardDeepSeek-V4-Pro by 11.446.234.8+11.4
- LiveCodeBench v6DeepSeek-V4-Pro by 7.592.585+7.5
- SWE-bench ProDeepSeek-V4-Pro by 4.755.450.7+4.7
- SWE-bench VerifiedDeepSeek-V4-Pro by 3.880.676.8+3.8
- SWE-bench MultilingualDeepSeek-V4-Pro by 3.276.273+3.2
- LMArena · WebDevDeepSeek-V4-Pro by 28 rating points14641436+28 rating
- SciCodeDeepSeek-V4-Pro by 1.850.849+1.8
AgentsDeepSeek-V4-Pro
- τ-Bench V3 · BankingDeepSeek-V4-Pro by 15.930.114.2+15.9
- GDPValDeepSeek-V4-Pro by 15.732.917.2+15.7
- AA ApexAgentsDeepSeek-V4-Pro by 12.824.311.5+12.8
- MCP AtlasDeepSeek-V4-Pro by 9.873.663.8+9.8
- BrowseCompDeepSeek-V4-Pro by 8.583.474.9+8.5
ReasoningDeepSeek-V4-Pro
- CritPtDeepSeek-V4-Pro by 9.812.93.1+9.8
- Humanity's Last ExamDeepSeek-V4-Pro by 6.837.530.7+6.8
- GPQA Diamondtie88.887.9tie
FactsDeepSeek-V4-Pro
- AA-Omniscience · Non-hallucinationKimi K2.5 by 28.45.934.3+28.4
- SimpleQA VerifiedDeepSeek-V4-Pro by 11.946.234.3+11.9
- AA-Omniscience · AccuracyDeepSeek-V4-Pro by 7.84335.2+7.8
- Vectara HHEM hallucination ratelower is betterDeepSeek-V4-Pro by 5.68.614.2+5.6
MathDeepSeek-V4-Pro
- HMMT Feb. 2026DeepSeek-V4-Pro by 8.195.287.1+8.1
- IMOAnswerBenchDeepSeek-V4-Pro by 889.881.8+8
- AIME 2026tie96.795.8tie
Following instructionsDeepSeek-V4-Pro
- IFBenchDeepSeek-V4-Pro by 6.376.570.2+6.3
Other results23 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.
- CyberGymDeepSeek-V4-Pro by 4283.341.3+42
- GDPval-AA v2DeepSeek-V4-Pro by 297 rating points13061009+297 rating
- Tool-DecathlonDeepSeek-V4-Pro by 2552.827.8+25
- ToolathlonDeepSeek-V4-Pro by 2451.827.8+24
- Terminal-Bench 2.0DeepSeek-V4-Pro by 17.167.950.8+17.1
- Artificial Analysis Coding IndexDeepSeek-V4-Pro by 12.659.446.8+12.6
- τ³-Bench BankingDeepSeek-V4-Pro by 11.82614.2+11.8
- HLE (with tools)DeepSeek-V4-Pro by 9.86050.2+9.8
- BrowseComp (w/ Ctx)DeepSeek-V4-Pro by 8.583.474.9+8.5
- AA IntelligenceDeepSeek-V4-Pro by 6.930.423.5+6.9
- NL2RepoDeepSeek-V4-Pro by 6.538.532+6.5
- AA Agentic IndexDeepSeek-V4-Pro by 627.721.7+6
- vectara_factual_consistencyDeepSeek-V4-Pro by 5.691.485.8+5.6
- Global-MMLU-LiteDeepSeek-V4-Pro by 5.389.384+5.3
- HMMT Nov. 2025DeepSeek-V4-Pro by 5.294.489.2+5.2
- vectara_answer_rateDeepSeek-V4-Pro by 597.292.2+5
- SWEBench Pro PublicDeepSeek-V4-Pro by 4.755.450.7+4.7
- LiveBenchDeepSeek-V4-Pro by 4.573.669.1+4.5
- vectara_avg_summary_lengthDeepSeek-V4-Pro by 41.8 rating points153.8112+41.8 rating
- AA-OmniscienceKimi K2.5 by 3.4-10.7-7.3+3.4
- StrongREJECTtie98.699.5tie
- MMLU-Protie87.587.1tie
- τ²-Bench Telecom (AA run)tie96.295.9tie
Questions people ask
Which is better, DeepSeek-V4-Pro or Kimi K2.5?
DeepSeek-V4-Pro wins six of the seven areas where both have results: coding, agents, reasoning, facts, math and following instructions. Kimi K2.5 wins long documents.
Which is better for coding?
DeepSeek-V4-Pro. It wins 8 of the 8 coding tests both models report; Kimi K2.5 wins none.
Which is cheaper?
DeepSeek-V4-Pro costs $0.43 per million input tokens and $0.87 per million output tokens; Kimi K2.5 costs $0.60 and $3.00. That makes DeepSeek-V4-Pro about 64% cheaper for the same work.
How do you compare the two?
We use the 48 benchmark tests both models have published scores on. The verdict counts the 25 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 23 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.