Hy3-preview vs Kimi K2.5
Wins 5 of 7 areas
Coding · Agents · Reasoning · Math · Long documents
Wins 2 of 7 areas
Facts · Following instructions
Hy3-preview is the stronger all-rounder.Kimi K2.5 is better at facts.
Scores updated · 29 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.
- AgentsCarrying out multi-step tasks on its own50Hy3-preview5 of 5 tests
- CodingWriting and fixing software51Hy3-preview5 of 7 tests · 1 tie
- ReasoningHard problems that need careful thinking30Hy3-preview3 of 3 tests
- MathCompetition and research-level math10Hy3-preview1 of 1 test
- Long documentsFinding answers in very long texts10Hy3-preview1 of 1 test
- FactsGetting facts right instead of making them up02Kimi K2.52 of 2 tests
- Following instructionsDoing exactly what it is asked01Kimi K2.51 of 1 test
Math, 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.
Hy3-preview costs 81% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Hy3-preview pulls ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+18.7points ahead
- Multi-step tasks using many real software toolsMCP Atlas+15.3points ahead
- Long professional tasks in banking, consulting and lawAA ApexAgents+14.1points ahead
Where Kimi K2.5 pulls ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+22.2points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+8.4points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+3.2points ahead
Every test, side by side
All 29 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingHy3-preview
- Terminal-Bench 2.1Hy3-preview by 18.764.445.7+18.7
- LMArena · WebDevHy3-preview by 72 rating points15081436+72 rating
- SWE-bench ProHy3-preview by 7.257.950.7+7.2
- Terminal-Bench HardKimi K2.5 by 331.834.8+3
- SWE-bench MultilingualHy3-preview by 2.875.873+2.8
- SWE-bench VerifiedHy3-preview by 1.27876.8+1.2
- SciCodetie48.649tie
AgentsHy3-preview
- MCP AtlasHy3-preview by 15.379.163.8+15.3
- AA ApexAgentsHy3-preview by 14.125.611.5+14.1
- GDPValHy3-preview by 11.128.317.2+11.1
- BrowseCompHy3-preview by 9.384.274.9+9.3
- τ-Bench V3 · BankingHy3-preview by 8.722.914.2+8.7
ReasoningHy3-preview
- Humanity's Last ExamHy3-preview by 2.833.530.7+2.8
- CritPtHy3-preview by 1.84.93.1+1.8
- GPQA DiamondHy3-preview by 1.889.787.9+1.8
FactsKimi K2.5
- AA-Omniscience · Non-hallucinationKimi K2.5 by 8.425.934.3+8.4
- AA-Omniscience · AccuracyKimi K2.5 by 3.23235.2+3.2
Following instructionsKimi K2.5
- IFBenchKimi K2.5 by 22.24870.2+22.2
Other results9 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)Kimi K2.5 by 28.467.595.9+28.4
- ToolathlonHy3-preview by 20.748.527.8+20.7
- DeepSearchQA (F1)Hy3-preview by 13.99177.1+13.9
- NL2RepoHy3-preview by 13.645.632+13.6
- Artificial Analysis Coding IndexHy3-preview by 1258.846.8+12
- AA-OmniscienceKimi K2.5 by 11.2-18.5-7.3+11.2
- AA Agentic IndexHy3-preview by 3.925.621.7+3.9
- WideSearchKimi K2.5 by 2.676.479+2.6
- AA IntelligenceHy3-preview by 1.825.323.5+1.8
Questions people ask
Which is better, Hy3-preview or Kimi K2.5?
Hy3-preview wins five of the seven areas where both have results: coding, agents, reasoning, math and long documents. Kimi K2.5 wins facts and following instructions.
Which is better for coding?
Hy3-preview. It wins 5 of the 7 coding tests both models report; Kimi K2.5 wins 1, and 1 is a tie.
Which is cheaper?
Hy3-preview costs $0.14 per million input tokens and $0.56 per million output tokens; Kimi K2.5 costs $0.60 and $3.00. That makes Hy3-preview about 81% cheaper for the same work.
How do you compare the two?
We use the 29 benchmark tests both models have published scores on. The verdict counts the 20 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 9 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.