GLM-5.2 vs GLM-5.3
Wins 0 of 7 areas
—
Wins 6 of 7 areas
Coding · Agents · Reasoning · Facts · Long documents · Following instructions
GLM-5.3 is the stronger all-rounder.
Scores updated · 43 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 software04GLM-5.34 of 5 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own04GLM-5.34 of 4 tests
- ReasoningHard problems that need careful thinking14GLM-5.34 of 5 tests
- FactsGetting facts right instead of making them up12GLM-5.32 of 3 tests
- Long documentsFinding answers in very long texts01GLM-5.31 of 1 test
- Following instructionsDoing exactly what it is asked01GLM-5.31 of 1 test
- MathCompetition and research-level math11Even1 each · 1 tie
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.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where GLM-5.3 pulls ahead
- Complex command-line tasks across many fieldsTerminal-Bench 4.0+40.9points ahead
- Customer-service tasks in a simulated bankτ-Bench V3 · Banking+15.7points ahead
- Real work tasks from 44 professionsGDPVal+13.8points ahead
Where GLM-5.2 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+3.3points ahead
- Fresh competition math problemsLiveBench · Mathematics+1.9points ahead
- Research-level physics problemsCritPt+1.8points ahead
Every test, side by side
All 43 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGLM-5.3
- LiveBench · Agentic CodingGLM-5.3 by 9.151.860.9+9.1
- SciCodeGLM-5.3 by 7.851.259+7.8
- Terminal-Bench 2.1GLM-5.3 by 677.983.9+6
- LMArena · WebDevGLM-5.3 by 19 rating points16031622+19 rating
- LiveBench · Codingtie79.779tie
AgentsGLM-5.3
- Terminal-Bench 4.0GLM-5.3 by 40.9141.9+40.9
- τ-Bench V3 · BankingGLM-5.3 by 15.734.650.3+15.7
- GDPValGLM-5.3 by 13.843.757.5+13.8
- AA IT-Bench SREGLM-5.3 by 3.442.746.1+3.4
ReasoningGLM-5.3
- SimpleBenchGLM-5.3 by 7.458.866.2+7.4
- LiveBench · ReasoningGLM-5.3 by 7.278.685.8+7.2
- GPQA DiamondGLM-5.3 by 2.289.591.7+2.2
- CritPtGLM-5.2 by 1.820.919.1+1.8
- Humanity's Last ExamGLM-5.3 by 1.241.142.3+1.2
FactsGLM-5.3
- AA-Omniscience · AccuracyGLM-5.3 by 9.624.333.9+9.6
- SimpleQA VerifiedGLM-5.3 by 6.834.241+6.8
- AA-Omniscience · Non-hallucinationGLM-5.2 by 3.373.770.4+3.3
MathEven
- FrontierMath Tiers 1-3 (v2)GLM-5.3 by 9.659.268.8+9.6
- LiveBench · MathematicsGLM-5.2 by 1.989.887.9+1.9
- FrontierMath Tier 4tie29.329.3tie
Following instructionsGLM-5.3
- LiveBench · Instruction FollowingGLM-5.3 by 762.369.3+7
Other results21 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.
- Program BenchGLM-5.2 by 44.763.719+44.7
- SWE-MarathonGLM-5.3 by 29.51342.5+29.5
- GDPval-AA v2GLM-5.3 by 261 rating points15081769+261 rating
- ToolathlonGLM-5.3 by 24.848.273+24.8
- Terminal-bench 3.0GLM-5.3 by 23.74.628.3+23.7
- DeepSWE 1.1GLM-5.3 by 22.94466.9+22.9
- DeepSWEGLM-5.3 by 20.746.266.9+20.7
- AA Agentic IndexGLM-5.3 by 1439.453.4+14
- Toolathlon VerifiedGLM-5.3 by 13.159.973+13.1
- AA IntelligenceGLM-5.3 by 11.133.744.8+11.1
- AA-OmniscienceGLM-5.3 by 9.94.414.3+9.9
- ProgramBench (Almost Solved)GLM-5.3 by 9.59.519+9.5
- NL2RepoGLM-5.3 by 9.148.958+9.1
- HLE (with tools)GLM-5.3 by 7.854.762.5+7.8
- CyberGymGLM-5.3 by 7.377.284.5+7.3
- Artificial Analysis Coding IndexGLM-5.3 by 668.874.8+6
- PostTrainBenchGLM-5.3 by 5.534.339.8+5.5
- Agents' Last ExamGLM-5.3 by 4.723.828.5+4.7
- Agents' Last Exam (ALE-CLI)GLM-5.3 by 4.723.828.5+4.7
- livebench_languageGLM-5.3 by 3.776.279.9+3.7
- livebench_data_analysisGLM-5.2 by 3.573.770.2+3.5
Questions people ask
Which is better, GLM-5.2 or GLM-5.3?
GLM-5.3 wins six of the seven areas where both have results: coding, agents, reasoning, facts, long documents and following instructions. GLM-5.2 wins none. They are level on math.
Which is better for coding?
GLM-5.3. It wins 4 of the 5 coding tests both models report; GLM-5.2 wins none, and 1 is a tie.
Which is cheaper?
GLM-5.2 costs $1.40 per million input tokens and $4.40 per million output tokens; GLM-5.3 costs $1.40 and $4.40. For a million tokens read plus a million written, they cost about the same.
How do you compare the two?
We use the 43 benchmark tests both models have published scores on. The verdict counts the 22 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 21 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.