The short answer
“GLM-4.5 Claude Code” can mean two different things: GLM-4.5 was evaluated using Claude Code, or you can configure GLM-4.5 as Claude Code’s model. Those claims are not interchangeable.
A July 28, 2025 article reports that Z.ai used Claude Code for a 52-task GLM-4.5 coding benchmark. Separately, Z.ai’s GLM-4.5 documentation says the model can be integrated into coding agents such as Claude Code. Z.ai’s current Claude Code guide documents a GLM Coding Plan setup and maps the default Claude Code model slots to GLM-5.3-Flash; its FAQ confirms that all plans support GLM-5.3 and GLM-5.3-Flash. Those plan documents do not establish whether GLM-4.5 is currently available through the Coding Plan. They also do not, by themselves, provide a verified GLM-4.5 production configuration for Claude Code. (Benchmark report; GLM-4.5 overview; Claude Code guide; Coding Plan FAQ)
What Z.ai reported about the Claude Code benchmark
According to Simon Willison’s account of Z.ai’s benchmark description, GLM-4.5 was compared with Claude-4-Sonnet, Kimi K2, and Qwen3-Coder across 52 tasks. The task areas included frontend development, tool development, data analysis, testing, and algorithm implementation. The quoted Z.ai results were a 53.9% win rate against Kimi K2 and an 80.8% success rate against Qwen3-Coder. Z.ai described GLM-4.5 as competitive with Claude-4-Sonnet, while noting that further optimization remained; the article gives no numeric result for that comparison. (Simon Willison’s report)
Treat those numbers as vendor-reported results, not proof that GLM-4.5 will outperform another model on your codebase. The article does not explain the scoring protocol behind the percentages or provide task-level results in its captured text. Its author also said they had not seen independent benchmarks at the time. That statement describes the evidence available in July 2025; it is not a claim about all evidence available now.
Most importantly, the benchmark says Claude Code was the evaluation environment. It is not, by itself, setup guidance for selecting GLM-4.5 as Claude Code’s backend.
Can you use GLM-4.5 in Claude Code through a production API?
Z.ai’s GLM-4.5 overview says the model can be integrated with Claude Code, but the cited Coding Plan instructions do not confirm a GLM-4.5 route for that plan. The current Claude Code guide maps the default model slots to GLM-5.3-Flash, and the FAQ confirms support for GLM-5.3 and GLM-5.3-Flash. Those statements confirm support for the named models; they do not establish that the FAQ is an exhaustive list or that GLM-4.5 is excluded. (GLM-4.5 overview; Claude Code guide; Coding Plan FAQ)
Z.ai’s Coding Plan documentation gives an Anthropic Messages endpoint, https://api.z.ai/api/anthropic, for that plan. Z.ai’s Quick Start also says newly created API keys for Individual Plans, Team Plans, and the standard model API are not interchangeable. That key guidance does not say that every previously created key is interchangeable or incompatible. The endpoint and key information alone do not confirm that a particular account can use GLM-4.5 through Claude Code on the Coding Plan. (Coding Plan endpoint guide; Z.ai Quick Start)
For a production decision, confirm with the provider that your specific account and API product support GLM-4.5 through the intended Claude Code integration. Ask for the exact model identifier, endpoint, authentication method, and configuration instructions. The available documentation supports the general integration statement, but the specific GLM-4.5 production route remains unconfirmed here. Do not assume that substituting a GLM-4.5 name into the current Coding Plan setup will work.
Once the provider confirms a route, validate it in two stages:
- Send a small authenticated request using the provider’s documented model identifier and endpoint. Confirm that it succeeds and identifies the expected model.
- In a disposable branch or test repository, ask Claude Code to make a small change, run the project’s tests, and report what it changed. Confirm that responses, repository edits, and tool calls all work as expected.
These are proposed checks, not tests performed for this article. A successful basic API response alone does not establish that a longer Claude Code agent workflow is reliable.
Is GLM-4.5 a good choice for your coding workflow?
Consider it for an evaluation if you can confirm a supported production route; do not choose it on benchmark percentages alone. The reported task categories make the results relevant to agentic coding, but the missing scoring detail and dated, vendor-attributed figures limit what they can predict for an individual team. Z.ai’s GLM-4.5 overview reports a 128K context length, but a model-page claim should not be mistaken for a guarantee about the limits or behavior of a particular hosted API plan. (Benchmark report; GLM-4.5 overview)
A practical comparison is to run the same representative tasks through GLM-4.5 and your current model. Include a bug fix, a multi-file change, a test-writing task, and a task that requires repository tools. Score each attempt on correctness, test results, how often you need to intervene, and end-to-end time. If API cost matters, compare actual billed cost per accepted task using your provider’s current prices and usage from your own runs.
| Your situation | What the evidence supports | Sensible next step |
|---|---|---|
| You want to know whether GLM-4.5 was benchmarked with Claude Code | Yes. A dated report describes Z.ai using Claude Code for a 52-task evaluation. |
Treat the reported percentages as vendor claims, not a guarantee for your repository. |
You want GLM-4.5 through Z.ai’s current GLM Coding Plan |
Z.ai’s guide documents Claude Code setup and default mappings to GLM-5.3-Flash; the FAQ confirms GLM-5.3 and GLM-5.3-Flash support. These excerpts do not establish whether GLM-4.5 is available through the plan. |
Confirm GLM-4.5 access, the model identifier, and plan terms with Z.ai before configuring it. |
| You want GLM-4.5 through another production API | Z.ai’s model overview says GLM-4.5 can integrate with Claude Code, but that does not verify another provider’s endpoint or service conditions. |
Obtain provider-specific model, protocol, authentication, and availability documentation. |
| You are choosing a coding model | The reported benchmark is a reason to test GLM-4.5, not enough to call it a universal winner. | Compare it on your actual tasks and reliability requirements before adoption. |
Bottom line
Z.ai reported using Claude Code to benchmark GLM-4.5, and its model overview says GLM-4.5 can integrate with Claude Code. Separately, Z.ai’s Coding Plan guide maps default model slots to GLM-5.3-Flash, and its FAQ confirms support for GLM-5.3 and GLM-5.3-Flash. Those Coding Plan statements do not establish whether GLM-4.5 is available through that plan or rule it out. Confirm the specific GLM-4.5 API route and account access with the provider before configuring Claude Code, then evaluate the full workflow on representative coding tasks. (Benchmark report; GLM-4.5 overview; Claude Code guide; Coding Plan FAQ; Z.ai Quick Start)
Sources
GLM-4.5: Reasoning, Coding, and Agentic Abilities
GLM-4.5 – Z.AI Developer Documentation
Claude Code – Z.AI Developer Documentation
FAQ – Z.AI Developer Documentation