Claude Sonnet 4.5 vs GLM-5.2
Wins 0 of 7 areas
—
Wins 7 of 7 areas
Coding · Agents · Reasoning · Facts · Math · Long documents · Following instructions
GLM-5.2 is the stronger all-rounder.
Scores updated · 32 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 software05GLM-5.25 of 5 tests
- ReasoningHard problems that need careful thinking05GLM-5.25 of 5 tests
- AgentsCarrying out multi-step tasks on its own04GLM-5.24 of 4 tests
- MathCompetition and research-level math03GLM-5.23 of 3 tests
- FactsGetting facts right instead of making them up12GLM-5.22 of 3 tests
- Long documentsFinding answers in very long texts01GLM-5.21 of 1 test
- Following instructionsDoing exactly what it is asked01GLM-5.21 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.
GLM-5.2 costs 68% 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 GLM-5.2 pulls ahead
- Unpublished advanced math problemsFrontierMath Tiers 1-3 (v2)+35.3points ahead
- Multi-step tasks using many real software toolsMCP Atlas+33points ahead
- Extremely hard research-level math problemsFrontierMath Tier 4+26.9points ahead
Where Claude Sonnet 4.5 pulls ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+8.6points ahead
Every test, side by side
All 32 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGLM-5.2
- Terminal-Bench 2.1GLM-5.2 by 22.155.877.9+22.1
- LMArena · WebDevGLM-5.2 by 209 rating points13941603+209 rating
- SWE-bench ProGLM-5.2 by 18.543.662.1+18.5
- Terminal-Bench HardGLM-5.2 by 15.235.650.8+15.2
- SciCodeGLM-5.2 by 5.545.751.2+5.5
AgentsGLM-5.2
- MCP AtlasGLM-5.2 by 3343.876.8+33
- GDPValGLM-5.2 by 23.220.543.7+23.2
- τ-Bench V3 · BankingGLM-5.2 by 10.124.534.6+10.1
- Terminal-Bench 4.0GLM-5.2 by 101+1
ReasoningGLM-5.2
- Humanity's Last ExamGLM-5.2 by 23.317.841.1+23.3
- CritPtGLM-5.2 by 19.81.120.9+19.8
- ARC-AGI-2GLM-5.2 by 9.213.622.8+9.2
- GPQA DiamondGLM-5.2 by 6.183.489.5+6.1
- SimpleBenchGLM-5.2 by 4.554.358.8+4.5
FactsGLM-5.2
- AA-Omniscience · Non-hallucinationGLM-5.2 by 22.850.973.7+22.8
- AA-Omniscience · AccuracyClaude Sonnet 4.5 by 8.632.924.3+8.6
- SimpleQA VerifiedGLM-5.2 by 3.530.734.2+3.5
MathGLM-5.2
- FrontierMath Tiers 1-3 (v2)GLM-5.2 by 35.323.959.2+35.3
- FrontierMath Tier 4GLM-5.2 by 26.92.429.3+26.9
- IMOAnswerBenchGLM-5.2 by 25.165.991+25.1
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.
- HLE (with tools)GLM-5.2 by 22.73254.7+22.7
- AA Agentic IndexGLM-5.2 by 21.917.539.4+21.9
- τ²-Bench Telecom (AA run)GLM-5.2 by 2178.199.1+21
- HMMT 2025GLM-5.2 by 19.874.694.4+19.8
- Artificial Analysis Coding IndexGLM-5.2 by 16.752.168.8+16.7
- ARC-AGI-1GLM-5.2 by 13.363.777+13.3
- AA IntelligenceGLM-5.2 by 1320.733.7+13
- HMMT Nov. 2025GLM-5.2 by 12.781.794.4+12.7
- ToolathlonGLM-5.2 by 7.24148.2+7.2
- AA-OmniscienceGLM-5.2 by 4.5-0.14.4+4.5
Questions people ask
Which is better, Claude Sonnet 4.5 or GLM-5.2?
GLM-5.2 wins all seven areas where both have results: coding, agents, reasoning, facts, math, long documents and following instructions. Claude Sonnet 4.5 wins none.
Which is better for coding?
GLM-5.2. It wins 5 of the 5 coding tests both models report; Claude Sonnet 4.5 wins none.
Which is cheaper?
Claude Sonnet 4.5 costs $3.00 per million input tokens and $15.00 per million output tokens; GLM-5.2 costs $1.40 and $4.40. That makes GLM-5.2 about 68% cheaper for the same work.
How do you compare the two?
We use the 32 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 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.