Kimi K2.6 vs Nemotron 3 Ultra 550B A55B
Wins 5 of 7 areas
Coding · Agents · Reasoning · Facts · Long documents
Wins 1 of 7 areas
Following instructions
Kimi K2.6 is the stronger all-rounder.Nemotron 3 Ultra 550B A55B is cheaper and better at following instructions.
Scores updated · 59 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 software80Kimi K2.68 of 9 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own40Kimi K2.64 of 5 tests · 1 tie
- ReasoningHard problems that need careful thinking40Kimi K2.64 of 4 tests
- FactsGetting facts right instead of making them up21Kimi K2.62 of 3 tests
- Long documentsFinding answers in very long texts10Kimi K2.61 of 1 test
- Following instructionsDoing exactly what it is asked02Nemotron 3 Ultra 550B A55B2 of 2 tests
- MathCompetition and research-level math22Even2 each
Long documents rests on a single test.
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.
Nemotron 3 Ultra 550B A55B costs 37% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Kimi K2.6 pulls ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+41.9points ahead
- Multi-step tasks using many real software toolsMCP Atlas+23.4points ahead
- Harvard-MIT high-school math contest problemsHMMT Feb. 2026+13.9points ahead
Where Nemotron 3 Ultra 550B A55B pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+10.8points ahead
- Follows detailed instructions exactlyLiveBench · Instruction Following+9.1points ahead
- Olympiad math problems with short answersIMOAnswerBench+6.3points ahead
Every test, side by side
All 59 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingKimi K2.6
- Terminal-Bench 2.1Kimi K2.6 by 1265.953.9+12
- SciCodeKimi K2.6 by 11.251.540.3+11.2
- SWE-bench VerifiedKimi K2.6 by 9.580.270.7+9.5
- SWE-bench MultilingualKimi K2.6 by 976.767.7+9
- SWE-bench ProKimi K2.6 by 8.958.649.7+8.9
- LiveBench · Agentic CodingKimi K2.6 by 8.246.938.7+8.2
- LiveBench · CodingKimi K2.6 by 7.978.670.7+7.9
- Terminal-Bench HardKimi K2.6 by 7.543.936.4+7.5
- LiveCodeBench v6tie89.689tie
AgentsKimi K2.6
- BrowseCompKimi K2.6 by 41.986.344.4+41.9
- MCP AtlasKimi K2.6 by 23.468.144.7+23.4
- τ-Bench V3 · BankingKimi K2.6 by 9.123.314.2+9.1
- GDPValKimi K2.6 by 1.226.925.7+1.2
- Terminal-Bench 4.0tie0.50.5tie
ReasoningKimi K2.6
- Humanity's Last ExamKimi K2.6 by 9.137.528.4+9.1
- CritPtKimi K2.6 by 4.983.1+4.9
- LiveBench · ReasoningKimi K2.6 by 4.779.474.7+4.7
- GPQA DiamondKimi K2.6 by 4.491.186.7+4.4
FactsKimi K2.6
- AA-Omniscience · Non-hallucinationNemotron 3 Ultra 550B A55B by 10.859.570.3+10.8
- AA-Omniscience · AccuracyKimi K2.6 by 1032.622.6+10
- SimpleQA VerifiedKimi K2.6 by 2.534.932.4+2.5
MathEven
- HMMT Feb. 2026Kimi K2.6 by 13.992.778.8+13.9
- IMOAnswerBenchNemotron 3 Ultra 550B A55B by 6.38692.3+6.3
- LiveBench · MathematicsNemotron 3 Ultra 550B A55B by 4.484.388.7+4.4
- AIME 2026Kimi K2.6 by 2.296.494.2+2.2
Following instructionsNemotron 3 Ultra 550B A55B
- LiveBench · Instruction FollowingNemotron 3 Ultra 550B A55B by 9.164.473.5+9.1
- IFBenchNemotron 3 Ultra 550B A55B by 5.47681.4+5.4
Other results31 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.
- Apex (Pass@1)Nemotron 3 Ultra 550B A55B by 60.82484.8+60.8
- HLE (with tools)Kimi K2.6 by 16.65437.4+16.6
- τ²-Bench Telecom (AA run)Kimi K2.6 by 12.695.983.3+12.6
- Artificial Analysis Coding IndexKimi K2.6 by 12.561.849.3+12.5
- SWEBench Pro PublicKimi K2.6 by 12.258.646.4+12.2
- Apex-Shortlist (with tools)Nemotron 3 Ultra 550B A55B by 11.673.284.8+11.6
- livebench_data_analysisKimi K2.6 by 10.665.154.5+10.6
- SciCode (subtask)Kimi K2.6 by 8.55243.5+8.5
- IFBench (prompt loose) (Tools-allowed)Nemotron 3 Ultra 550B A55B by 873.781.7+8
- Finance Agent v2Kimi K2.6 by 7.444.937.5+7.4
- Vals.ai Financial Agent 1.1 - without web searchNemotron 3 Ultra 550B A55B by 6.15460.1+6.1
- TauBench V3 - AirlineKimi K2.6 by 5.885.880+5.8
- AA-OmniscienceKimi K2.6 by 5.75.3-0.4+5.7
- CritPt (no tools)Kimi K2.6 by 5.79.13.4+5.7
- TauBench V3 - RetailNemotron 3 Ultra 550B A55B by 5.582.988.4+5.5
- Vals.ai Financial Agent 1.1 - with web searchKimi K2.6 by 5.158.853.7+5.1
- livebench_languageKimi K2.6 by 4.375.170.8+4.3
- GDPval-AA v2Kimi K2.6 by 42 rating points11901148+42 rating
- AA IntelligenceKimi K2.6 by 4.12722.9+4.1
- Global-MMLU-LiteKimi K2.6 by 2.888.485.6+2.8
- Apex-Shortlist (no tools)Kimi K2.6 by 2.577.474.9+2.5
- τ³-Bench BankingKimi K2.6 by 2.220.618.4+2.2
- TauBench V3 - AverageKimi K2.6 by 2.172.470.3+2.1
- IOI 2025Kimi K2.6 by 20.3 rating points585564.7+20.3 rating
- MMLU-ProX (avg en/de/fr/es/it/ja/zh/hi/pt/ko) (Tools-allowed)Kimi K2.6 by 28583+2
- IMOAnswerBench (with tools)Kimi K2.6 by 1.493.792.3+1.4
- StrongREJECTKimi K2.6 by 1.199.898.7+1.1
- AA Agentic Indextie22.121.7tie
- ProfBench (Search)tie5656.4tie
- MMLU-Protie87.186.8tie
- PinchBenchtie90.290tie
Questions people ask
Which is better, Kimi K2.6 or Nemotron 3 Ultra 550B A55B?
Kimi K2.6 wins five of the seven areas where both have results: coding, agents, reasoning, facts and long documents. Nemotron 3 Ultra 550B A55B wins following instructions, and costs 37% less. They are level on math.
Which is better for coding?
Kimi K2.6. It wins 8 of the 9 coding tests both models report; Nemotron 3 Ultra 550B A55B wins none, and 1 is a tie.
Which is cheaper?
Kimi K2.6 costs $0.95 per million input tokens and $4.00 per million output tokens; Nemotron 3 Ultra 550B A55B costs $0.60 and $2.50. That makes Nemotron 3 Ultra 550B A55B about 37% cheaper for the same work.
How do you compare the two?
We use the 59 benchmark tests both models have published scores on. The verdict counts the 28 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 31 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.