GLM 4.6 — Benchmarks

Benchmark scores for GLM 4.6 aggregated from public leaderboards (8 boards), with how it ranks among open models and how it compares with its nearest neighbours. See hardware requirements for what you need to run it.

llmrun Composite Score →

103.1

90% range 98.4–107.9 · from 6 boards

Coding96.090% range 81–111
Agents104.390% range 99–109
MathNot enough data yet
Science99.890% range 79–121
ReasoningNot enough data yet

GLM 4.6 reasoning benchmarks

Multi-step logic and problem-solving — can the model think through a hard problem rather than recall an answer.

CritPt
CritPt (Critical Points) tests graduate-level physics problem-solving on research-style questions spanning mechanics, electromagnetism, quantum mechanics, and more. It was created by physicists to probe whether a model can reason through multi-step physics rather than pattern-match textbook answers.
1.1%
1.1%
median 1.1%open leader 26.6%
Overall best: GPT 5.6 Sol Max 32.3%
52nd percentile · #33 of 66 open · #159 of 219 all
#33 of 66
#159 of 219
Lab-reported
Epoch AI · Sep 2025

GLM 4.6 coding benchmarks

Writing, editing, and fixing real code — measured by whether the resulting program actually runs and passes tests.

SWE-bench Verified
SWE-bench Verified tests whether a model can resolve real GitHub issues from popular open-source Python projects, scored on the official swebench.com leaderboard as the percentage of human-validated issues actually fixed. It is the headline measure of practical, agentic software-engineering ability — where open-weight models like Qwen3-Coder, GLM-4.6, Kimi K2 and DeepSWE are now competitive with the frontier.
68.2%
68.2%
median 62.2%open leader 75.8%
Overall best: live-SWE-agent + Claude 4.5 Opus medium (20251101) 79.2%
68th percentile · #7 of 20 open · #50 of 162 all
#7 of 20
#50 of 162
Independent
SWE-bench · Sep 2025
Terminal-Bench
Terminal-Bench measures whether a model can complete real, end-to-end tasks in a command-line environment — running commands, editing files, and chaining steps — making it an agentic test of practical software skill.
24.5%
24.5%
median 31.7%open leader 52.4%
Overall best: GPT 5.5 (unspecified) 84.7%
20th percentile · #13 of 16 open · #50 of 57 all
#13 of 16
#50 of 57
Lab-reported
Epoch AI · Sep 2025
SWE-bench Bash Only
SWE-bench Bash Only runs the SWE-bench Verified issues through a minimal, single-tool bash agent — no specialised scaffolding — so the score reflects the model's own agentic coding ability rather than the harness around it. A cleaner, harder read on raw software-engineering skill.
55.4%
55.4%
median 55.4%open leader 63.4%
Overall best: Claude 4.5 Opus (high reasoning) 76.8%
56th percentile · #5 of 9 open · #32 of 48 all
#5 of 9
#32 of 48
Independent
SWE-bench · Dec 2025
SciCode
SciCode asks a model to write real scientific-computing code — implementing multi-step numerical methods drawn from physics, chemistry, biology, and math research — and checks the output against reference solutions. It was built by a team of scientists and ML researchers to test coding ability on genuine research problems, not typical software tasks.
38.4%
38.4%
median 38.7%open leader 60.9%
Overall best: Claude Opus 5.5 Max 66.9%
48th percentile · #35 of 66 open · #171 of 207 all
#35 of 66
#171 of 207
Lab-reported
Epoch AI · Sep 2025
ALE-Bench
ALE-Bench, from Sakana AI and AtCoder, tests long-horizon algorithmic optimisation using problems drawn from AtCoder Heuristic Contests, where there is no single correct answer and solutions are scored on quality. Results are reported as an AtCoder-style performance rating, where higher is better.
340.8
340.8
median 694.1open leader 1524.5
Overall best: GPT 6 Astra Max 2951.3
12th percentile · #38 of 43 open · #110 of 122 all
#38 of 43
#110 of 122
Lab-reported
Epoch AI · Sep 2025

Other results (1)· leaderboard-only, not part of the llmrun Score

LMArena WebDev
Code Arena (formerly WebDev Arena) is Arena's head-to-head coding leaderboard: two models each build a real web app or UI from the same prompt, and people vote blind on which result is better. The rating shown is a relative Bradley-Terry score, not a percentage — compare it only to other models on this board.
1341.1
1341.1
median 1411.4open leader 1655.2
Overall best: Claude Opus 5.5 (Max) 1807.3
24th percentile · #40 of 52 open · #112 of 138 all
#40 of 52
#112 of 138
Arena
Arena · Oct 2026

GLM 4.6 instruction following benchmarks

How precisely a model obeys explicit, verifiable instructions — format, length, and content constraints — and, more broadly, how much people prefer its responses head-to-head in blind human voting.

LMArena Text
Arena (arena.ai, formerly LMArena / Chatbot Arena) ranks models by millions of blind, head-to-head human votes on real prompts. The Arena Score shown here is a Bradley-Terry rating computed with Arena's default style control, so higher means people preferred that model's answers more often — it's a relative rating, not a percentage, so compare it only to other models on this board.
1424.1
1424.1
median 1321.5open leader 1487.1
Overall best: Gemini 4 Argon (High) 1523.9
84th percentile · #31 of 187 open · #122 of 405 all
#31 of 187
#122 of 405
Arena
Arena · Oct 2026

Score vs VRAM in the ≤256 GB tier

GLM 4.6
Other models
Best Score for its VRAM
x: VRAM (GB, log scale) · y: llmrun Score

#32 of 84 that fit ≤256 GB

Each dot is a model at the best quant that fits ≤256 GB. The line joins the models with the best Score for their VRAM; a dot below it is beaten by a model that needs no more memory.

Full ≤256 GB ranking →

GLM 4.6 vs its neighbours

Same boards for the nearest open models per llmrun Score. The highest score in each row is marked.

Board	GLM 4.6this model	Qwen3.8 27Bbest in classQ4_K_M · 18 GB	MiniMax M2nearest by ScoreQ4_K_M · 140 GB	Qwen3 Coder 480B A35B Instructnearest by ScoreQ4_K_M · 291 GB	DeepSeek R1 0528nearest by ScoreQ4_K_M · 412 GB
llmrun Score	103.1	122.1	104.0	103.9	103.7
CritPt	1.1%	5.4%	—	—	—
SWE-bench Verified	68.2%	—	61.0%	69.6%	—
Terminal-Bench	24.5%	—	30.0%	27.2%	—
SWE-bench Bash Only	55.4%	—	61.0%	55.4%	—
SciCode	38.4%	46.6%	—	—	—
ALE-Bench	340.8	—	—	461.4	804.1
LMArena Text	1424.1	1437.8	1343.3	1387.6	1421.4
Compare head to head		Compare	Compare	Compare	Compare

Scores aggregated from public benchmark sources (Epoch AI: 4 boards, Arena: 2 boards, SWE-bench: 2 boards) — each linked from the benchmark pages. llmrun does not run these benchmarks. Percentiles and ranks are among open models in the llmrun catalogue; the median and leader marks refer to the same field.