Zhipu · tested in Taskade Genesis
GLM 5.2 in Taskade
GLM 5.2 from Zhipu runs on the Max and Enterprise plans. We last tested it on Sep 19, 2026 in TSK-1, our app-building test.
- Status
- Available
- Plan
- Max and Enterprise
- Where to pick it
- More models or All Models
- Your own key
- Together AI
- Last test
- Context window
- 1Mtokens
- Max output
- 131Ktokens
- Released
- Where it runs
- 19services
TSK-1 scorecard for the GLM family
TSK Score
GLM family, across real apps made in Taskade Genesis
- Interfaceready to share
- Emerging
- Taskcompletes your task
- Strong
- Memorykeeps your data
- Emerging
- Adapthandles follow-up edits
- Strong
The model that said no in July, and a steady finisher in September.
MoreLess
It built a live win-rate dashboard and a three-page sign-up app with 31 of 32 questions word for word.
Published scores
Benchmark scores that Zhipu and test sites published for GLM 5.2. Taskade did not re-run them. Lab-reported marks a score from Zhipu itself.
| Benchmark | Score | Source |
|---|---|---|
| SWE-Bench Proresolve rate · Z.aiLab-reported | 62.1% | Z.aiLab-reported |
| Terminal-Benchsuccess rate · v2.1 · Claude Code · Z.aiLab-reported | 82.7% | Z.aiLab-reported |
| MCP Atlaspublic subset · Z.aiLab-reported | 76.8 | Z.aiLab-reported |
| GPQA Diamondaccuracy · Z.aiLab-reported | 91.2% | Z.aiLab-reported |
| Humanity's Last Examaccuracy · text-only subset · Z.aiLab-reported | 40.5% | Z.aiLab-reported |
More published scores (14)
| Benchmark | Score | Source |
|---|---|---|
| FrontierSWEdominance · Z.aiLab-reported | 74.4 | Z.aiLab-reported |
| Humanity's Last Examaccuracy · with tools · text-only subset · Z.aiLab-reported | 54.7% | Z.aiLab-reported |
| CritPtaccuracy · Z.aiLab-reported | 20.9% | Z.aiLab-reported |
| AIMEaccuracy · v2026 · Z.aiLab-reported | 99.2% | Z.aiLab-reported |
| HMMTaccuracy · vNovember 2025 · Z.aiLab-reported | 94.4% | Z.aiLab-reported |
| HMMTaccuracy · vFebruary 2026 · Z.aiLab-reported | 92.5% | Z.aiLab-reported |
| IMOAnswerBenchaccuracy · Z.aiLab-reported | 91% | Z.aiLab-reported |
| NL2Reporesolve rate · Z.aiLab-reported | 48.9% | Z.aiLab-reported |
| DeepSWEresolve rate · Z.aiLab-reported | 46.2% | Z.aiLab-reported |
| Program BenchZ.aiLab-reported | 63.7 | Z.aiLab-reported |
| Terminal-Benchsuccess rate · v2.1 · Terminus 2 · Z.aiLab-reported | 81% | Z.aiLab-reported |
| PostTrainBenchZ.aiLab-reported | 34.3 | Z.aiLab-reported |
| SWE Marathonresolve rate · Z.aiLab-reported | 13% | Z.aiLab-reported |
| Tool-DecathlonZ.aiLab-reported | 48.2 | Z.aiLab-reported |
LMArena scores
| Benchmark | Score | Source |
|---|---|---|
| Text arena±4 · 49,820 votes · Rank 32 of 416 · LMArena | 1,476 | LMArena |
| WebDev arena±6 · 16,602 votes · Rank 29 of 145 · LMArena | 1,604 | LMArena |
Scores from the LMArena leaderboard dataset (CC BY 4.0). People vote between two answers, and the votes set the score. Taskade picked the Overall category and the default setting of each model.
Scores collected by models.dev (MIT). Each source links the page that published the score.
Versions tested
Open a version to read its findings.
GLM 5.2Tested Sep 20265 findings
The model that said no in July, and a steady finisher in September. It built a live win-rate dashboard and a three-page sign-up app with 31 of 32 questions word for word.
Thought about adding a sign-in screen, decided the brief had not asked for one, and offered "Add Login" as a suggestion instead. The only model to push back rather than quietly add it.
Its app saved real data for the first time. A small styling issue is the one thing still open.
Complete and working, but its sample matches carried no result, so the dashboard showed a 0% win rate until a real match was logged.
Built a two-page match tracker whose dashboard showed a 58% win rate over 12 sample matches: 7 wins and 5 losses.
Built a three-page client sign-up app with a 37-field submissions table and 31 of 32 questions word for word.
Model facts
The token limits are the most the model accepts. Taskade can run it with lower limits.
- Last updated
- Input types
- Text
- Output types
- Text
- Reasoning
- Yes
- Tool calling
- Yes
- File attachments
- No
- Structured output
- Yes
- Open weights
- Yes
- Reasoning effort
- HighMax
Published specs from the model maker, via models.dev (MIT).
Run GLM 5.2 with your own key
Taskade runs it for you. You can also pick it with your own key in a custom agent.
| Key provider | Plan |
|---|---|
| Together AI | Business and up |
Where it runs
Services outside Taskade that serve GLM 5.2, from models.dev. Each link opens in a new tab. In Taskade, your own key works only through the key providers on this page.
How to choose GLM 5.2
- In the builder chat, open the model menu and choose More models. Then pick the model.
- In the agent editor, pick the model under All Models.
- If your model menu does not show the model, Taskade uses TSK-Auto instead.
Where GLM ranks
TSK Score out of 100.
Compare GLM
Other Zhipu models
FAQ
Which Taskade plans include GLM 5.2?
GLM 5.2 runs in Taskade on the Max and Enterprise plans. If your model menu does not show it, Taskade uses TSK-Auto instead.
Has Taskade tested GLM 5.2?
Yes. We last tested GLM 5.2 on Sep 19, 2026 in the TSK-1 Benchmark. The model that said no in July, and a steady finisher in September. It built a live win-rate dashboard and a three-page sign-up app with 31 of 32 questions word for word.
How do I choose GLM 5.2 in Taskade?
In the builder chat, open the model menu and choose More models. Then pick GLM 5.2. In the agent editor, pick it under All Models.
Can I use my own API key for GLM 5.2?
Yes. In a custom agent, you can pick GLM 5.2 with your own key from Together AI (Business and up). Your provider bills the model use to your own account.
Who makes GLM 5.2?
Zhipu makes GLM 5.2. Taskade runs it on the Max and Enterprise plans, beside frontier models from other top AI labs.