Carbon-X vs Hy3

Carbon-X

A DNA specialist cousin that another agent calls for variant-effect and sequence questions, with DNA kept on the machine. Hugging Face's Q4_K_M GGUF needs about 12 GB and Q8_0 about 19 GB, on llama.cpp b11742 or later. It isn't clinically validated, it isn't trained for tool calls or reasoning, and the card contradicts its own GGUF repository on llama.cpp support.

Hy3

Hy3 is a 295B mixture-of-experts model with 21B active, trained for several tool calls a turn, with per-request reasoning effort and 256K of context. It needs an 8-GPU server for FP8 or about 200 GB for a 4-bit build. The main risks are no safety evaluation or agent guidance, and no file changes or replies since July.

Carbon-X

1. Quickest start is llama-server -hf HuggingFaceBio/Carbon-X-GGUF:Q8_0 -ngl 99 -c 40960 --jinja --chat-template-kwargs '{"enable_thinking": false}' --temp 0.6 --top-p 0.95 --top-k 20 --min-p 0, on llama.cpp b11742 or later
2. With vLLM or Transformers pass trust_remote_code, and expect the tokenizer to fetch Qwen/Qwen3-4B-Base at a pinned revision on first load
3. Keep enable_thinking false and the card's sampling (temperature 0.6, top_p 0.95, top_k 20), the settings every reported result used
4. Keep DNA in files and pass sequences by name, compute positions and base changes in code and send them as context, the card says the model miscounts long sequences
5. Only A, C, G and T form 6-mers, so filter N and other IUPAC codes first or they become a single oov token
6. Strip the dna tags from answers if you need bare sequence, llama.cpp keeps them in the text
7. Cap max_tokens for next-base questions, long DNA generation is slow and degrades

Hy3

1. Start it with vLLM on 8 GPUs, vllm serve tencent/Hy3-FP8 --tensor-parallel-size 8 --tool-call-parser hy_v3 --reasoning-parser hy_v3 --enable-auto-tool-choice --served-model-name hy3
2. In SGLang use --tool-call-parser hunyuan --reasoning-parser hunyuan, as the card shows
3. Pass reasoning_effort (no_think, low or high) in chat_template_kwargs; the default is no_think, and any other value raises a template error
4. With the vLLM reasoning parser the chain of thought comes back in message.reasoning and the answer in message.content
5. Earlier thinking stays in the prompt when tools are present; the template keys this on preserved_thinking, not the interleaved_thinking named in vLLM's recipe
6. Use temperature 0.9 and top_p 1.0, the card's recommended sampling
7. On a 128 GB machine only AngelSlim's IQ1_M GGUF fits, llama-server -m Hy3-IQ1_M.gguf -ctk q8_0 -ctv q8_0 -fa on -c 65536, with llama.cpp b11768 or later; at 89.4 GB × 1.2 it sits right at the 107.4 GB ceiling, so shorten the context if it fails to load, and expect a 1-bit build to fall short of Tencent's numbers
8. Without the hardware, call it through Featherless AI or DeepInfra via Hugging Face's inference providers