GLM 5.2
Jun 2026
GLM 5.2Compare inon
Benchmarks of GLM 5.288
BenchmarkByScoreRank
Overall1
Knowledge7
Reasoning5
Math17
Science3
Coding9
Agents8
Writing20
Health & law5
Documents2
Legacy · saturated or no longer reported11
No benchmark matches
Rankingat max effort?
Alternatives?
ModelScoreCostPer taskWhat changes
CheaperGPT-5.6 Solxhigh×5.9-8%$1.18OpenGLM 5.3max×7.2+56%$2.01Reasoning cannot be disabled+3
CheaperMiMo v2.6 ProthinkingOpen×3.7-90%$0.13Memory (4-bit): 452 GB → 615 GB+4
CheaperMuse Spark 1.1highsame-7%$1.25 / $4.25Only n=1 · Meta / Chat Completions+1CheaperMuse Spark 1.3highsame-7%$1.25 / $4.25Only auto tool_choice is supported+5ModelScoreCostPer taskWhat changes
GPT-5.6 Solxhigh×5.9-8%$1.18MiMo v2.6 ProthinkingOpen×3.7-90%$0.13Memory (4-bit): 452 GB → 615 GB+4Muse Spark 1.1highsame-7%$1.25 / $4.25Only n=1 · Meta / Chat Completions+1Muse Spark 1.3highsame-7%$1.25 / $4.25Only auto tool_choice is supported+5
DeepSeek V4 Pro 0813highOpensame-8%$1.32 / $3.96Memory (4-bit): 452 GB → 990 GB+5ModelScoreCostPer taskWhat changes
GPT-5.6 Solxhigh×5.9-8%$1.18GPT-6 Astrahigh×8.3+34%$1.73Monitoring can return HTTP 403+2
Claude Opus 5.5high×9+42%$1.82No forced tool use+5GLM 5.3maxOpen×7.2+56%$2.01Reasoning cannot be disabled+3Claude Opus 5high×8.1+182%$3.61ModelScoreCostPer taskWhat changes
GLM 5.3maxOpen×7.2+56%$2.01Reasoning cannot be disabled+3MiMo v2.6 ProthinkingOpen×3.7-90%$0.13Memory (4-bit): 452 GB → 615 GB+4DeepSeek V4 Pro 0813highOpensame-8%$1.32 / $3.96Memory (4-bit): 452 GB → 990 GB+5
Kimi K2.6unstatedOpensame-20%$0.95 / $4Context: 1M → 256K+1
MiniMax M3unstatedOpensame-76%$0.30 / $1.20Context: 1M → 524K+2ModelScoreCostPer taskWhat changes
GLM 5.3maxOpen×7.2+56%$2.01Reasoning cannot be disabled+3GLM 5.1unstatedOpen-363%—$1.40 / $4.40Context: 1M → 203KGLM 5.3 FlashXOpenn/a—$0.37 / $1.25Memory (4-bit): not verifiedGLM 5.3 FlashmaxOpenn/a—$0.28Defaults to max · Self-hosted+4GLM 5V Turbothinkingn/a—$1.20 / $4Context: 1M → 203KGLM 5 Turbothinkingn/a—$1.20 / $4Context: 1M → 203KGLM 5thinkingOpenn/a—$1 / $3.20Context: 1M → 203KGLM 4.7 FlashthinkingOpenn/a—$0.061 / $0.40Context: 1M → 200KAll Z.ai models →Run it yourself
estimateFits on✗ 8 GB (phone, laptop)✗ 16 GB✗ 24 GB GPU✗ 48 GB GPU✗ 80 GB GPU
LicenceMITfree for commercial use
History
10 versionsBenchmarks
all categoriesChat & Knowledge
Humanity's Last Exam41.1#41
Arena Text · overall1,476#24
Arena Text · Chinese1,521.9#21
Arena Text · English1,474.8#23
Arena Text · French1,478.1#27
Arena Text · German1,470.9#20
Arena Text · instruction following1,465.9#26
Arena Text · Japanese1,448.5#22
Arena Text · Korean1,437.9#19
Arena Text · non-English1,458.5#23
Arena Text · Polish1,457.1#25
Arena Text · Russian1,467#25
Arena Text · Spanish1,477.6#12
AA Omniscience Index4.4#55
SimpleQA Verified v134.2#51
CritPt (physics)20.9#27
AA Intelligence Index 4.3.233.7#49
AA Omniscience · Accuracy24.3#122
AA-LCR (long context) 1.178.3#63
GDP.pdf · AA all-pass10.4#60
MLCR-AA12.8#43
FrontierMath Tiers 1–3 (v2)59.2#36
FrontierMath Tier 4 (v2)29.3#30
Chess Puzzles22#48
Arena Text · overall · October 22, 20261,469.8#23
Arena Text · overall · October 15, 20261,470.4#23
GPQA Diamond89.5#41
GPQA Diamond · Epoch91.9#20
OTIS Mock AIME 2024 202586.4#62
Coding
DeepSWE 1.143.8#29
WebDev Arena1,604#24
DeepSWE 141.5#5
SciCode51.2#48
SWE-bench Pro (edition unstated)lab62.1
DeepSWE (edition unstated)lab46.2
SciCode Verified pass@1 n=6lab88.1
Surge Human Eval on Codelab3.4
WebDev Arena · October 23, 20261,604.3#25
WebDev Arena · October 19, 20261,604.2#24
Terminal-Bench 2.177.9#26
Agents & Tool Use
AutomationBench-AA28.4#51
GDPval-AA (work tasks) 2.11,373#43
τ³-Banking · AA34.6#26
ITBench SRE42.7#13
APEX-Agents33.7#7
EnterpriseOps-Gym-AA42.7#16
AA-Briefcase 1.1 · Elo1,227#44
EBR Bench9.5#21
AutomationBenchlab28.4
Finance Agent v2lab49.7
Harvey Legal Agent Benchmarklab7.1
Cybenchlab73
Finch FinWorkBenchlab61
B3 Attack Resistancelab88.1
Harmful cyber request refusal meanlab90
KORA aggregatelab1.4
τ²-bench · AA99.1#1
Writing & Language
Arena Text · creative writing1,454#21
Lech Mazur Short-Story0.4#20
IFBench (instructions)73.3#30
FAQ
How much does GLM 5.2 cost?›
$1.40 per 1M input tokens and $4.40 per 1M output tokens; cached input $0.26 per 1M; cheapest via Ambient at $0.60 / $2. A typical benchmark task costs about $1.28 at max effort.
Is GLM 5.2 free to use?›
There is no free API; the cheapest paid price is $0.60 per 1M input tokens. The weights are open, so you can run it yourself.
How good is GLM 5.2, and what is it best at?›
#41 of 336 in Chat & Knowledge, #56 of 134 in Coding, #51 of 135 in Agents & Tool Use, #15 of 197 in Writing & Language, #60 of 135 in Vision & Documents. Strongest at: Agents & work tasks, Tool calling, Reasoning, Science, Long context, Knowledge, Instructions.
Is GLM 5.2 better than GLM 5.1?›
On Lech Mazur Short-Story, GLM 5.2 scores 0.4 vs -1.1 — higher.
How can I use GLM 5.2?›
Through the API of Alibaba, Ambient, Azure AI Foundry, BaseTen, Cloudflare and 14 more.
What is the context window of GLM 5.2?›
1M tokens, with up to 131K output tokens.
When was GLM 5.2 released? Is there a newer version?›
Released 2026-06-13. It has been superseded by GLM 5.3.
Is GLM 5.2 open source?›
Yes, the weights are open (MIT licence).
Can I run GLM 5.2 locally?›
The weights are open and can be self-hosted; hardware needs are not confirmed yet.
Can GLM 5.2 read images?›
No, it accepts text only.
What is GLM 5.2?›
GLM 5.2 is an open-weights language model from Z.ai, released Jun 2026, the successor to GLM 5.1.
What are the best alternatives to GLM 5.2?›
GPT-5.6 Sol (cheaper), GLM 5.3 (open weights).