GLM 4.7
Z.ai's GLM-4.7 via the Vercel AI Gateway: a coding, reasoning, and agentic model with 200k context and function calling. Served only by zero-data-retention providers. Does not support web search or image input.
You always get the exact model you pick — we never silently route you to another.
About GLM 4.7
GLM-4.7 is Zhipu AI's coding-specialist built for developers who want frontier-grade software engineering without frontier-grade pricing. Its sparse Mixture-of-Experts design activates only about 3B parameters per pass from a 400B pool, which is how it scores 73.8% on SWE-bench Verified — a result that puts it above much larger proprietary models — while staying at $0.40 per million input tokens on OpenRouter. Users consistently call it the best sub-70B model they have run for real development work: implementing features, hunting bugs, and writing documentation across a 200K-token context window. Agentic and tool-use tasks are a particular strength; testers report it outhandles Qwen3 and similar open-weight alternatives on multi-step workflows. The honest caveat is that pure abstract reasoning is not its home turf — one evaluator found "disappointing results" on LiveBench logic puzzles, and it remains a step behind Claude Sonnet on frontier reasoning tasks. If your work is code-first and cost-conscious, GLM-4.7 earns its place; if you need deep chain-of-thought over ambiguous logic problems, you may want something else.
Best for
- Software engineering and code review — high SWE-bench scores translate directly to practical feature implementation and bug fixing
- Multi-step agentic workflows — reliable tool use across extended, multi-turn interactions with three configurable thinking modes
- Mathematical problem-solving — 95.7% on AIME 2025 makes it a strong pick for competitive math and quantitative reasoning
- Large-codebase tasks — the 200K input and 128K output window lets you load entire projects and generate substantial outputs in a single pass
- Bilingual (English and Chinese) development and research — native support for both languages without quality trade-offs
Specifications
| Provider | Zhipu AI |
|---|---|
| Released | 2025-12 |
| Intelligence | Medium |
| Speed | Medium |
| Context window | 200,000 tokens |
| Max output | 128,000 tokens |
| Knowledge cutoff | Not specified; model released December 2025 |
| Input price | $0.60 / 1M tokens |
| Output price | $2.20 / 1M tokens |
| Request cost | 1 premium request |
| Plan tier | Premium |
| Input | Text |
| Output | Text |
| Features | Function calling supported, Served through the Vercel AI Gateway (zai/glm-4.7), Zero data retention and no prompt training |
| Model ID | glm-4p7 |
Frequently asked questions
Via OpenRouter and Z.ai's official API, input runs $0.39–$0.40 per million tokens and output $1.75 per million. A free variant, GLM-4.7-Flash, is also available, and context caching can cut cached input costs to roughly $0.11 per million tokens.
GLM-4.7 supports 200,000 input tokens and up to 128,000 output tokens — large enough to process entire codebases and generate lengthy completions in a single call.
Coding and software engineering tasks are its signature strength, with a 73.8% SWE-bench Verified score. Mathematical reasoning (95.7% AIME 2025) and agentic tool use are also standouts.
Abstract logic puzzles and complex reasoning chains are a relative weakness — testers noted disappointing LiveBench reasoning results. Quality can also degrade at the extreme ends of the context window.
It is a significant step up from GLM-4.6, most visibly on Humanity's Last Exam (+12.4 percentage points) and overall coding benchmarks, with architecture improvements focused on reasoning and agentic performance.
The series has three members: the standard GLM-4.7, the lightweight free GLM-4.7-Flash, and the high-speed paid GLM-4.7-FlashX. A GLM-4.7 Reasoning variant with extended thinking is also available.