Gemma 4 31B vs GLM-5.2
Wins 1 of 7 areas
Following instructions
Wins 6 of 7 areas
Coding · Agents · Reasoning · Facts · Math · Long documents
GLM-5.2 is the stronger all-rounder.Gemma 4 31B is cheaper and better at following instructions.
Scores updated · 30 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
- AgentsCarrying out multi-step tasks on its own05GLM-5.25 of 5 tests
- ReasoningHard problems that need careful thinking03GLM-5.23 of 3 tests
- FactsGetting facts right instead of making them up03GLM-5.23 of 3 tests
- MathCompetition and research-level math03GLM-5.23 of 3 tests
- Long documentsFinding answers in very long texts01GLM-5.21 of 1 test
- Following instructionsDoing exactly what it is asked10Gemma 4 31B1 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.
Gemma 4 31B costs 90% 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
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+58.7points ahead
- Real work tasks from 44 professionsGDPVal+37.6points ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+34.5points ahead
Where Gemma 4 31B pulls ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+2.3points ahead
Every test, side by side
All 30 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 34.543.477.9+34.5
- SWE-bench ProGLM-5.2 by 26.435.762.1+26.4
- LMArena · WebDevGLM-5.2 by 238 rating points13651603+238 rating
- Terminal-Bench HardGLM-5.2 by 14.436.450.8+14.4
- SciCodeGLM-5.2 by 5.745.551.2+5.7
AgentsGLM-5.2
- GDPValGLM-5.2 by 37.66.143.7+37.6
- τ-Bench V3 · BankingGLM-5.2 by 19.814.834.6+19.8
- MCP AtlasGLM-5.2 by 19.657.276.8+19.6
- AA IT-Bench SREGLM-5.2 by 5.437.342.7+5.4
- Terminal-Bench 4.0GLM-5.2 by 101+1
ReasoningGLM-5.2
- CritPtGLM-5.2 by 19.51.420.9+19.5
- Humanity's Last ExamGLM-5.2 by 17.523.641.1+17.5
- GPQA DiamondGLM-5.2 by 3.885.789.5+3.8
FactsGLM-5.2
- AA-Omniscience · Non-hallucinationGLM-5.2 by 58.71573.7+58.7
- SimpleQA VerifiedGLM-5.2 by 23.810.434.2+23.8
- AA-Omniscience · AccuracyGLM-5.2 by 4.32024.3+4.3
MathGLM-5.2
- IMOAnswerBenchGLM-5.2 by 16.574.591+16.5
- HMMT Feb. 2026GLM-5.2 by 15.377.292.5+15.3
- AIME 2026GLM-5.2 by 1089.299.2+10
Following instructionsGemma 4 31B
- IFBenchGemma 4 31B by 2.375.673.3+2.3
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.
- AA-OmniscienceGLM-5.2 by 52.3-47.94.4+52.3
- τ²-Bench Telecom (AA run)GLM-5.2 by 39.259.999.1+39.2
- NL2RepoGLM-5.2 by 33.415.548.9+33.4
- AA Agentic IndexGLM-5.2 by 32.76.739.4+32.7
- HLE (with tools)GLM-5.2 by 28.226.554.7+28.2
- Tool-DecathlonGLM-5.2 by 2721.248.2+27
- Artificial Analysis Coding IndexGLM-5.2 by 25.443.468.8+25.4
- AA IntelligenceGLM-5.2 by 1914.733.7+19
- HMMT Nov. 2025GLM-5.2 by 6.987.594.4+6.9
Questions people ask
Which is better, Gemma 4 31B or GLM-5.2?
GLM-5.2 wins six of the seven areas where both have results: coding, agents, reasoning, facts, math and long documents. Gemma 4 31B wins following instructions, and costs 90% less.
Which is better for coding?
GLM-5.2. It wins 5 of the 5 coding tests both models report; Gemma 4 31B wins none.
Which is cheaper?
Gemma 4 31B costs $0.17 per million input tokens and $0.40 per million output tokens; GLM-5.2 costs $1.40 and $4.40. That makes Gemma 4 31B about 90% cheaper for the same work.
How do you compare the two?
We use the 30 benchmark tests both models have published scores on. The verdict counts the 21 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.