The Open-Source AI Stack
RSS
All models

Models · glm-4

GLM-4.6

Open Zhipu AI · 2025-09-30 · MIT

September 30 2025 refresh, 357B MoE with 200K context (up from 128K in 4.5) and 128K maximum output. Zhipu reports a 27 percent coding improvement over 4.5 and parity with Claude Sonnet 4 on 8 public benchmarks. MIT licensed.

Cost

$0.60 / Mtok input
$2.20 / Mtok output

· as of 2026-05-21

source ↗

Speed

30.7 tok/sec output
1860 ms TTFT

· as of 2026-05-21

source ↗

Architecture

tokens in Embedding vocab not disclosed × N layers Attention (not disclosed) Position encoding not disclosed context 128,000 tokens MoE Router ? experts total · ? active per token Output projection tokens out
Schema-generated from data/models.yaml. Every label is auditable against the model's sources.

Specs

Architecture
moe
Total params
357B
Active params
32B
Context window
128K tokens
Attention
unknown
Position encoding
unknown
Post-training
sft, rlhf
OSI-approved
yes
Data released
no
Training code
not released

Benchmarks

Each score carries the date it was published; we never infer or interpolate missing scores.

General reasoning

MMLU-Pro 78.4 as of 2026-05-21 source ↗
GPQA-Diamond 63.2 as of 2026-05-21 source ↗

Code

LiveCodeBench 56.1 as of 2026-05-21 source ↗

Math

AIME 2025 44.3 as of 2026-05-21 source ↗

Available quantizations

GGUF llama.cpp's container; the common local format, k-quants from Q2 to Q8. runs on llama.cpp, Ollama
AWQ Activation-aware 4-bit weight quantization for GPU serving. runs on vLLM, SGLang
GPTQ Post-training 4-bit weight quantization for GPU serving. runs on vLLM, SGLang, Transformers
EXL2 ExLlamaV2's variable-bitrate format for consumer GPUs. runs on ExLlamaV2
MLX Apple MLX 4/8-bit layout for Apple silicon. runs on Apple MLX
FP8 8-bit float, frequently a native release on Hopper / Blackwell GPUs. runs on vLLM, SGLang, TensorRT-LLM

Verified via the Hugging Face model tree ↗. Community quantizations change over time; the families shown are those with published weights at audit time.

Notable innovations

  • · 200K context (up from 128K)
  • · 30 percent token efficiency gain per Zhipu
  • · Parity with Claude Sonnet 4 claimed on 8 benchmarks

Lineage

200K context and a 27 percent coding bump over GLM-4.5 per Zhipu.

Derived from

GLM-4.5 2025-07-28

Sources