GLM-4.7-Flash vs Qwen3.8-Flash-Next

GLM-4.7-Flash

A small MIT-licensed mixture-of-experts model with about 3B active parameters, for local agent and coding work with several tool calls per turn and optional thinking. It runs in Ollama at about 23 GB and in nearly every common runtime. The main risk is neglect, with no changes or replies since early February and no safety results.

Qwen3.8-Flash-Next

A large sparse cousin with 6B active parameters for agentic coding and multimodal work, previewing Qwen's next architecture. It needs a 128 GB Mac for Ollama's MLX build or about 144 GB for the 4-bit GGUF, and Qwen Cloud serves it. The main risk is a licence that needs a separate deal for hosted-model and coding or office assistant businesses.

GLM-4.7-Flash

1. On our reference machine start it with vllm serve zai-org/GLM-4.7-Flash --max-model-len 65536 --tool-call-parser glm47 --reasoning-parser glm45 --enable-auto-tool-choice --served-model-name glm-4.7-flash
2. Or run ollama run glm-4.7-flash for the 19 GB q4_K_M build
3. Without the glm47 and glm45 parsers, tool calls and reasoning come back as plain text
4. Use a released vLLM (0.32.0 or later) or SGLang (0.5.22 or later); the card's nightly and pull-request builds date from January 2026
5. Pass chat_template_kwargs: {"enable_thinking": false} for short tasks, since thinking is on by default
6. Set clear_thinking: false to keep earlier reasoning in multi-turn agent work, as Z.ai does for τ²-Bench and Terminal Bench
7. Set the context length yourself, since config.json allows 202,752 tokens and tokenizer_config.json says 128,000

Qwen3.8-Flash-Next

1. On a 128 GB Mac start it with ollama run qwen3.8-flash-next:125b-mlx (105 GB, NVFP4), on Linux the smallest Ollama build is qwen3.8-flash-next:125b-a6b-q4_K_M at 120 GB, which won't fit a 107.4 GB GPU share
2. On a server, vllm serve Qwen/Qwen3.8-Flash-Next-FP8 --tensor-parallel-size 4 --max-model-len 262144 --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder
3. Use the qwen3_coder tool parser and the qwen3 reasoning parser, as Qwen's GitHub README does
4. Thinking is on by default, pass chat_template_kwargs with enable_thinking false to skip it, and read the reasoning from reasoning_content, even though SGLang's cookbook says thinking can't be turned off
5. reasoning_effort takes xhigh (default), medium or low, and any other value makes the template raise an error
6. Send at most one system message and put it first, since the template raises an error otherwise
7. Sampling per the card is temperature 1.0 and top_p 0.95 when thinking, temperature 0.7, top_p 0.8 and presence_penalty 1.5 without
8. vLLM keeps the 51B n-gram table in host memory on 80 GB GPUs by default, per its recipe
9. Validate JSON output yourself or use the runtime's guided decoding, since structured output isn't documented
10. Read the Qwen Community License before serving it to others as an API or inside a coding or office assistant product