GLM-5.3-Flash vs Kimi K2.5
Wins 5 of 6 areas
Coding · Agents · Reasoning · Images and charts · Long documents
Wins 1 of 6 areas
Facts
GLM-5.3-Flash is the stronger all-rounder.Kimi K2.5 is better at facts.
Scores updated · 28 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.
- ReasoningHard problems that need careful thinking40GLM-5.3-Flash4 of 4 tests
- CodingWriting and fixing software31GLM-5.3-Flash3 of 4 tests
- AgentsCarrying out multi-step tasks on its own31GLM-5.3-Flash3 of 4 tests
- Images and chartsUnderstanding pictures, charts and video20GLM-5.3-Flash2 of 2 tests
- Long documentsFinding answers in very long texts10GLM-5.3-Flash1 of 1 test
- FactsGetting facts right instead of making them up12Kimi K2.52 of 3 tests
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.
GLM-5.3-Flash costs 82% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where GLM-5.3-Flash pulls ahead
- Abstract visual puzzles that people can solveARC-AGI-2+54points ahead
- Real work tasks from 44 professionsGDPVal+40.2points ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+38.6points ahead
Where Kimi K2.5 pulls ahead
- Completes tasks by operating a computer desktopOSWorld-Verified+48.5points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+7.7points ahead
- Recent programming contest problemsLiveCodeBench v6+4.4points ahead
Every test, side by side
All 28 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGLM-5.3-Flash
- Terminal-Bench 2.1GLM-5.3-Flash by 38.684.345.7+38.6
- LMArena · WebDevGLM-5.3-Flash by 173 rating points16091436+173 rating
- LiveCodeBench v6Kimi K2.5 by 4.480.685+4.4
- SciCodeGLM-5.3-Flash by 2.651.649+2.6
AgentsGLM-5.3-Flash
- OSWorld-VerifiedKimi K2.5 by 48.514.863.3+48.5
- GDPValGLM-5.3-Flash by 40.257.217+40.2
- τ-Bench V3 · BankingGLM-5.3-Flash by 3347.214.2+33
- BrowseCompGLM-5.3-Flash by 1.17674.9+1.1
ReasoningGLM-5.3-Flash
- ARC-AGI-2GLM-5.3-Flash by 5465.811.8+54
- CritPtGLM-5.3-Flash by 12.315.43.1+12.3
- Humanity's Last ExamGLM-5.3-Flash by 9.239.930.7+9.2
- GPQA DiamondGLM-5.3-Flash by 3.391.287.9+3.3
FactsKimi K2.5
- AA-Omniscience · Non-hallucinationGLM-5.3-Flash by 38.172.434.3+38.1
- AA-Omniscience · AccuracyKimi K2.5 by 7.727.535.2+7.7
- SimpleQA VerifiedKimi K2.5 by 2.33234.3+2.3
Images and chartsGLM-5.3-Flash
- CharXiv (RQ)GLM-5.3-Flash by 11.989.477.5+11.9
- LMArena · VisionGLM-5.3-Flash by 26 rating points12951269+26 rating
Other results10 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.
- ToolathlonGLM-5.3-Flash by 50.678.427.8+50.6
- AA Agentic IndexGLM-5.3-Flash by 29.551.221.7+29.5
- ARC-AGI-1GLM-5.3-Flash by 25.79165.3+25.7
- Artificial Analysis Coding IndexGLM-5.3-Flash by 24.771.546.8+24.7
- NL2RepoGLM-5.3-Flash by 24.356.332+24.3
- AA IntelligenceGLM-5.3-Flash by 18.341.823.5+18.3
- BabyVisionGLM-5.3-Flash by 16.953.436.5+16.9
- AA-OmniscienceGLM-5.3-Flash by 14.87.5-7.3+14.8
- DeepSearchQA (F1)GLM-5.3-Flash by 7.284.377.1+7.2
- MMVUtie80.580.4tie
Questions people ask
Which is better, GLM-5.3-Flash or Kimi K2.5?
GLM-5.3-Flash wins five of the six areas where both have results: coding, agents, reasoning, images and charts and long documents. Kimi K2.5 wins facts.
Which is better for coding?
GLM-5.3-Flash. It wins 3 of the 4 coding tests both models report; Kimi K2.5 wins 1.
Which is cheaper?
GLM-5.3-Flash costs $0.15 per million input tokens and $0.50 per million output tokens; Kimi K2.5 costs $0.60 and $3.00. That makes GLM-5.3-Flash about 82% cheaper for the same work.
How do you compare the two?
We use the 28 benchmark tests both models have published scores on. The verdict counts the 18 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 10 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.