Model
GLM-4.7 NO.37
Efficient open coding workhorse for everyday dev tasks.
Zhipu AI · GLM · Open · Open weights
Specification
13 fields
Fig.01 — GLM — Zhipu AI
CHINA
- Context window
- 200Ktokens
- Max output
- 131Ktokens
- Input price
- $0.60/ 1M
- Output price
- $2.20/ 1M
- Cached input
- $0.11/ 1M
- Throughput
- —
- Class
- Open
- Modalities
- Text
- Weights
- Open
- Parameters
- 355B
- License
- MIT
- Knowledge cutoff
- Oct 2025
- Released
- Dec 2025
Overview
GLM-4.7 is Zhipu's cheaper, faster open-weight coding model, widely used as the value tier of the GLM Coding Plan and as a Claude Code-compatible backend. It handles routine multi-file coding, tool use, and reasoning capably at a fraction of flagship cost, but steps down from GLM-5.2 on the hardest long-horizon tasks. MIT-licensed and self-hostable.
Strengths
- Very low cost for capable coding
- Faster and lighter than GLM-5.2
- ~200K context for sizable repos
- MIT open weights, self-hostable
Trade-offs
- Below the flagship on complex multi-file work
- Text-only, no vision
- Agentic reliability well below the frontier
- Superseded by GLM-5.2 on hard tasks
Fit
Best for
- Budget coding assistants
- Routine refactors and bug fixes
- High-volume dev automation
- Self-hosted coding backends
Not ideal for
- Hardest long-horizon agentic coding
- Workloads needing 1M-token context
Capabilities
Normalized 0—100
Ten axes, normalized 0–100 and scored the same way across the whole catalog — so a 78 here means what a 78 means anywhere else on the bench.
Fig.02 — Capability radar
0 — 100
- Reasoning
- 74
- Coding
- 70
- Math
- 80
- Writing
- 76
- Knowledge
- 78
- Speed
- 80
- Agentic
- 66
- Vision
- 10
- Multilingual
- 80
- Long Context
- 74
Benchmarks
04 results
Public results, with independent third-party runs marked. Bars normalize percentages against 100 and Elo ratings against a 1500 ceiling.
- SWE-bench VerifiedIndependentCoding · SWE-bench
- 73.8%
- AIME 2025Math · Vendor-reported
- 95.7%
- GPQA DiamondIndependentReasoning · Independent aggregators
- 85.7%
- LiveCodeBenchIndependentCoding · LiveCodeBench
- 84.9%
Alternatives
03 comparable
Models in roughly the same class — the ones worth weighing against this record.
Index
Record 37 of 41