GLM-4.7

GLM-4.7 is the latest release in Z.ai's GLM family of language models, positioned as a step forward from GLM-4.6 for real-world development workloads. The model is designed around three core capabilities: high-quality code generation and editing, reliable tool use for agent-style workflows, and consistent multi-turn reasoning across extended conversations. According to its Cerebras Inference Cloud announcement, GLM-4.7 was built to combine this frontier-level intelligence with the throughput advantages of wafer-scale inference hardware, targeting developer use cases such as iterative coding sessions, tool-driven automation, and complex problem-solving dialogues that require sustained reasoning over many turns.

Within the open-weight landscape, GLM-4.7 is presented as a top-performing option for advanced developer benchmarks, with the announcement noting it leads DeepSeek-V3.2 across workloads including SWEbench, τ²bench, and LiveCodeBench, reflecting measurable gains in software engineering, agentic task completion, and competitive programming. Practical improvements over the previous generation show up as more accurate coding solutions, cleaner code structure, and stronger multilingual output, while remaining stable over long, iterative development loops. These qualities make GLM-4.7 well suited for teams seeking a self-hostable foundation for code assistants, agent pipelines, and reasoning-heavy applications without depending on closed proprietary APIs.

Merge Gatewayzai/glm-4.7glm

Open provider page Provider docs

Quick Info

Provider: Merge Gateway
Model key: zai/glm-4.7
Release date: Dec 22, 2025
Last updated: Dec 22, 2025
Knowledge cutoff: 2025-04
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $0.60
Output token cost: $2.20

Limits

Output tokens: 131,072 tokens
Context window: 200,000 tokens

Transparent token rates

Compare GLM-4.7 pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about GLM-4.7

Amazon Bedrock

Official sourceAnnouncement

Z.ai released GLM-4.7 on December 22, 2025, positioned as a step forward over GLM-4.6 for agentic coding and real development workflows. The model adds a "thinking before acting" execution pattern usable inside Claude Code, Kilo Code, Cline, and Roo Code, and improves "Vibe Coding" UI quality for webpages and slides. On coding benchmarks, GLM-4.7 reaches 73.8% on SWE-bench (+5.8), 66.7% on SWE-bench Multilingual (+12.9), 41% on Terminal Bench 2.0 (+16.5), and 84.9% on LiveCodeBench-v6. Reasoning gains include 42.8% on HLE with tools (+12.4), 85.7% on GPQA-Diamond, 95.7% on AIME 2025, and stronger τ²-Bench tool use.

Amazon Bedrock

CoverageAnalysis

Zhipu AI released GLM-4.7 on December 22, 2025 as an open-weight large language model built for deep mathematical reasoning, multi-file software engineering, and stable agentic orchestration. A third-party technical analysis describes the model as built on roughly 400 billion parameters with a 200,000-token context window and a maximum output capacity of 128,000 tokens, delivering inference at about 55 tokens per second for real-time development use. The article frames GLM-4.7 as a meaningful step up from GLM-4.6 across reasoning, mathematics, and coding workloads. Benchmark figures cited in the same analysis show GLM-4.7 scoring 42.8% on the Humanitarian Language Evaluation with tools, 95.7% on AIME 2025, 85.7% on GPQA-Diamond, 73.8% on SWE-bench Verified, and 84.9% on LiveCodeBench-v6, placing it at the open-weight frontier. The piece notes a 38% improvement over GLM-4.6 on HLE and ranks GLM-4.7 ahead of DeepSeek-V3.2 and MiMo-V2-Flash on SWE-bench Verified, supporting Zhipu AI's positioning of GLM-4.7 as a competitive open-weight coding model.

Amazon Bedrock

CoverageRelease Notes

Z.ai's December 27, 2025 release confirms GLM-4.7 shipped on December 22 as the latest GLM family iteration targeting multi-step production development environments. It emphasizes stability across long task cycles, frequent tool use, and "think-then-act" execution within Claude Code, Cline, Roo Code, TRAE, and Kilo Code frameworks. Z.ai reports GLM-4.7 ranks sixth in WebDev and first among open models, with measurable gains over GLM-4.6 in task completion and behavioral consistency across 100 real Claude Code-based programming tasks covering frontend, backend, and instruction-following scenarios.

Amazon Bedrock

CoverageBenchmark

A launch analysis from llm-stats confirms Zhipu AI released GLM-4.7 on December 22, 2025 as the company's latest flagship foundation model focused on complex reasoning, agentic coding, and long-context generation. The piece details a Mixture-of-Experts architecture optimized for compute efficiency, paired with a 200,000-token input context window and a 128,000-token output capacity that lets the model generate entire software frameworks in a single pass. The same report highlights GLM-4.7's "Vibe Coding" capabilities for frontend development and competitive positioning against proprietary systems like Claude 3.5 Sonnet. It positions the model as currently dominating open-source benchmarks, with pricing economics that the authors argue make it viable for production deployments alongside the larger context and output limits that distinguish it from competing frontier models.

Amazon Bedrock

Coverage

Z.ai's December 24, 2025 Business Wire announcement frames GLM-4.7 as an open-source release built on GLM-4.6 with strengthened support for coding workflows, complex reasoning, and agent-style execution. The release stresses consistency in long, multi-step tasks and stable behavior when interacting with external tools in production settings. The release also highlights gains in conversational, writing, and role-playing output, positioning GLM-4.7 as a more coherent open-source system. It reiterates the WebDev ranking of sixth overall and first among open models, reinforcing the model's positioning for real engineering use.