InternVL3.5-GPT-OSS-20B-A4B-Preview vs Qwopus3.8-27B-Flash-V2

InternVL3.5-GPT-OSS-20B-A4B-Preview

An image and video reader on gpt-oss-20b with an optional thinking mode, a preview untouched for over a year. It needs about 51 GB in BF16, since OpenGVLab publishes no quantised build, and runs in Transformers and vLLM. The main risk is a chat template in the Transformers-format repository that doesn't match its tokenizer, with no safety results.

Qwopus3.8-27B-Flash-V2

A Qwen3.8-27B cousin tuned to think less per turn in long agent loops, with an MTP head for speculative decoding in llama.cpp. It fits 24 GB with Jackrong's 4-bit GGUF. The risks are one person's experimental release with no safety work, training traces taken from Claude and GPT, and a tool format that differs from Qwen's.

InternVL3.5-GPT-OSS-20B-A4B-Preview

1. Serve with vllm serve OpenGVLab/InternVL3_5-GPT-OSS-20B-A4B-Preview-HF --max-model-len 32768, which is our command; the card gives none for vLLM
2. Check the rendered prompt before relying on it, since the shipped template's <|im_start|> markers aren't tokens in the gpt-oss tokenizer; the GitHub-format repository's harmony template is the alternative
3. Use Transformers 4.55.0 or later, as the card requires for this size
4. For thinking, send the card's thinking system prompt and parse the reasoning out of the <think> tags, with temperature 0.6 and sampling on as the card advises
5. Keep inputs within 32K tokens, the length it was trained at, whatever the config says

Qwopus3.8-27B-Flash-V2

1. Quickest start is llama-server -hf Jackrong/Qwopus3.8-27B-Flash-V2-GGUF:Q4_K_M --jinja, about 17 GB, and add mmproj-F32.gguf for images
2. The card's tests used Q5_K_M with the bundled MTP head and --spec-type draft-mtp in llama.cpp
3. The template asks for tool calls as JSON inside tool_call tags, so use a Hermes-style parser (vLLM's hermes), not Qwen3.8's qwen3_coder
4. For coding the card recommends enable_thinking true, reasoning_effort xhigh and temperature 0.3, which you must pass yourself, since generation_config.json says 1.0
5. reasoning_effort takes xhigh (default), medium or low, and any other value makes the template raise an error
6. Earlier reasoning stays in the context unless preserve_thinking is false
7. Cap max_tokens, since a report says it loops more than V1