GLM-5.1 vs Qwen3.5 397B A17B
Wins 5 of 7 areas
Coding · Agents · Reasoning · Facts · Math
Wins 2 of 7 areas
Long documents · Following instructions
GLM-5.1 is the stronger all-rounder.Qwen3.5 397B A17B is cheaper and better at long documents.
Both rank among the ten best models we track in following instructions.
Scores updated · 61 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 software60GLM-5.16 of 8 tests · 2 ties
- AgentsCarrying out multi-step tasks on its own40GLM-5.14 of 5 tests · 1 tie
- MathCompetition and research-level math31GLM-5.13 of 4 tests
- ReasoningHard problems that need careful thinking21GLM-5.12 of 3 tests
- FactsGetting facts right instead of making them up21GLM-5.12 of 3 tests
- Long documentsFinding answers in very long texts01Qwen3.5 397B A17B1 of 1 test
- Following instructionsDoing exactly what it is asked01Qwen3.5 397B A17B1 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.
Qwen3.5 397B A17B costs 27% 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-5.1 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+59points ahead
- Real work tasks from 44 professionsGDPVal+16.2points ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+10.5points ahead
Where Qwen3.5 397B A17B pulls ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+7.1points ahead
- Harvard-MIT high-school math contest problemsHMMT Feb. 2026+5.3points ahead
- Reasons across sets of long documentsAA-LCR+3.6points ahead
Every test, side by side
All 61 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGLM-5.1
- LMArena · WebDevGLM-5.1 by 108 rating points15081400+108 rating
- Terminal-Bench 2.1GLM-5.1 by 10.561.851.3+10.5
- SWE-bench ProGLM-5.1 by 7.558.450.9+7.5
- SWE-bench MultilingualGLM-5.1 by 473.369.3+4
- Terminal-Bench HardGLM-5.1 by 2.343.240.9+2.3
- LiveCodeBench v6GLM-5.1 by 2.185.783.6+2.1
- SWE-bench Verifiedtie76.276.4tie
- SciCodetie44.844.8tie
AgentsGLM-5.1
- GDPValGLM-5.1 by 16.23114.8+16.2
- BrowseCompGLM-5.1 by 10.379.369+10.3
- AA IT-Bench SREGLM-5.1 by 6.240.334.1+6.2
- Terminal-Bench 4.0GLM-5.1 by 220+2
- τ-Bench V3 · Bankingtie13.613.4tie
ReasoningGLM-5.1
- CritPtGLM-5.1 by 2.94.61.7+2.9
- GPQA DiamondQwen3.5 397B A17B by 2.586.889.3+2.5
- Humanity's Last ExamGLM-5.1 by 1.130.129+1.1
FactsGLM-5.1
- AA-Omniscience · Non-hallucinationGLM-5.1 by 5970.111.1+59
- SimpleQA VerifiedGLM-5.1 by 83426+8
- AA-Omniscience · AccuracyQwen3.5 397B A17B by 7.123.730.8+7.1
MathGLM-5.1
- FrontierMath Tiers 1-3 (v2)GLM-5.1 by 5.636.831.2+5.6
- HMMT Feb. 2026Qwen3.5 397B A17B by 5.382.687.9+5.3
- AIME 2026GLM-5.1 by 495.391.3+4
- IMOAnswerBenchGLM-5.1 by 2.983.880.9+2.9
Long documentsQwen3.5 397B A17B
- AA-LCRQwen3.5 397B A17B by 3.673.777.3+3.6
Following instructionsQwen3.5 397B A17B
- IFBenchQwen3.5 397B A17B by 2.576.378.8+2.5
Other results36 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-5.1 by 31.60.8-30.8+31.6
- Apex-Shortlist (with tools)GLM-5.1 by 18.67960.4+18.6
- Terminal-Bench 2.0GLM-5.1 by 16.56952.5+16.5
- AA Agentic IndexGLM-5.1 by 14.625.210.6+14.6
- AgentWorldBench - SWEQwen3.5 397B A17B by 12.352.164.4+12.3
- NL2RepoGLM-5.1 by 10.542.732.2+10.5
- Apex-Shortlist (no tools)GLM-5.1 by 9.771.161.4+9.7
- TauBench V3 - AirlineGLM-5.1 by 8.58576.5+8.5
- AgentWorldBench - SearchQwen3.5 397B A17B by 8.322.530.8+8.3
- AgentWorldBench - TerminalQwen3.5 397B A17B by 847.355.3+8
- AA IntelligenceGLM-5.1 by 7.726.118.4+7.7
- Artificial Analysis Coding IndexGLM-5.1 by 7.655.848.2+7.6
- ProfBench (Search)Qwen3.5 397B A17B by 74653+7
- IMOAnswerBench (with tools)GLM-5.1 by 6.691.184.5+6.6
- PinchBenchQwen3.5 397B A17B by 5.481.286.6+5.4
- TauBench V3 - RetailQwen3.5 397B A17B by 4.484.188.5+4.4
- AgentWorldBench - AndroidGLM-5.1 by 4.259.154.9+4.2
- HLE (with tools)GLM-5.1 by 452.348.3+4
- AgentWorldBench - OverallQwen3.5 397B A17B by 3.451.354.7+3.4
- AgentWorldBench - WebGLM-5.1 by 2.951.548.5+2.9
- ToolathlonGLM-5.1 by 2.440.738.3+2.4
- τ²-Bench Telecom (AA run)GLM-5.1 by 2.197.795.6+2.1
- AgentWorldBench - OSQwen3.5 397B A17B by 1.859.160.9+1.8
- MMLU-ProQwen3.5 397B A17B by 1.88687.8+1.8
- Vals.ai Financial Agent 1.1 - with web searchGLM-5.1 by 1.760.759+1.7
- IFBench (prompt loose) (Tools-allowed)Qwen3.5 397B A17B by 1.676.678.2+1.6
- IOI 2025GLM-5.1 by 15.2 rating points456.5441.3+15.2 rating
- CritPt (no tools)GLM-5.1 by 1.33.72.4+1.3
- HMMT 2025GLM-5.1 by 1.39492.7+1.3
- HMMT Nov. 2025GLM-5.1 by 1.39492.7+1.3
- TauBench V3 - AverageQwen3.5 397B A17B by 1.369.771+1.3
- Vals.ai Financial Agent 1.1 - without web searchQwen3.5 397B A17B by 1.160.261.3+1.1
- GPQA (unspecified)Qwen3.5 397B A17B by 186.187.1+1
- AgentWorldBench - MCPtie67.668.3tie
- MMLU-ProX (avg en/de/fr/es/it/ja/zh/hi/pt/ko) (Tools-allowed)tie85.886.4tie
- SciCode (subtask)tie47.748tie
Questions people ask
Which is better, GLM-5.1 or Qwen3.5 397B A17B?
GLM-5.1 wins five of the seven areas where both have results: coding, agents, reasoning, facts and math. Qwen3.5 397B A17B wins long documents and following instructions, and costs 27% less.
Which is better for coding?
GLM-5.1. It wins 6 of the 8 coding tests both models report; Qwen3.5 397B A17B wins none, and 2 are ties.
Which is cheaper?
GLM-5.1 costs $1.38 per million input tokens and $4.40 per million output tokens; Qwen3.5 397B A17B costs $0.60 and $3.60. That makes Qwen3.5 397B A17B about 27% cheaper for the same work.
How do you compare the two?
We use the 61 benchmark tests both models have published scores on. The verdict counts the 25 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 36 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.