GLM 4.7 Flash AI Evaluation | LayerLens - Benchmark Results & Performance

Model

GLM 4.7 Flash by z-ai

text->text

Free credits included. No credit card required.

As a 30B-class SOTA model, GLM-4.7-Flash offers a new option that balances performance and efficiency. It is further optimized for agentic coding use cases, strengthening coding capabilities, long-horizon task planning, and tool collaboration, and has achieved leading performance among open-source models of the same size on several current public benchmark leaderboards.

Evaluations

across 10 benchmarks

Latency

28.8s

Context length

200k

tokens

Cost

$0 · $0

input · output per 1M tokens

Key takeaways

The GLM 4.7 Flash is a 30B-class Mixture of Experts (MoE) model optimized for agentic coding use cases, balancing performance and efficiency.
The model demonstrates strong performance in mathematical reasoning across various domains (Algebra, Number Theory, Counting/Probability), achieving nearly 90% accuracy on the MATH-500 dataset.
Key capabilities include robust arithmetic, logical reasoning, and effective adherence to stringent formatting rules in mathematical contexts.
Limitations include occasional errors in complex geometry (especially diagram interpretation), highly abstract math problems, and advanced SQL operations (e.g., intricate window functions, temporal calculations, complex JSONB structures).
The model shows foundational SQL capabilities but struggles significantly with advanced features, achieving only 17.58% accuracy on the BIRD-CRITIC SQL Truthfulness Benchmark due to difficulties with complex data manipulation, temporal queries, and intricate window functions.

Technical specifications

Context length: 200k tokens
Max output: —
Architecture: MoE
Modalities: text->text
Parameters: ~30B
Open weights: Yes
Licence: MIT
Regions: China
Vendor: z-ai
Provider model key: z-ai/glm-4.7-flash