Carbon-X

# Fix list: Carbon-X

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/carbon-x, the October 2026 research run, assessed 23 October 2026. Grade C, 61 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on Carbon-X: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Risk, 65 out of 100, up to 10.8 more on the total

Why it scored 65: File safety 15, safetensors and GGUF only, but the documented loading path runs the repository's tokenizer.py through trust_remote_code. Provenance 15, the files are in HuggingFaceBio, which Hugging Face's GitHub repository links, with a checksum on every file. Safety evaluation 10, Hugging Face reports its own dual-use check on sequence editing in the launch post, without numbers or a harness. Agent guidance 5, general cautions only. Licence risk 15, plain Apache 2.0. Security programme 5, Hugging Face's security.txt is a reporting route, and public advisories couldn't be checked.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-risk):

What could go wrong when you run it, and what the publisher has done about it. A higher score means less risk.

- 0 to 25, file safety. 25 for weights in safetensors or GGUF only, 15 when pickle files (.bin, .pt, .pth) sit beside them, 0 for pickle only. 10 off when the documented way to load it or to build its prompts runs code from the repository (`trust_remote_code`, or the publisher's scripts in place of a template).
- 0 to 15, provenance. 10 when the files come from the publisher's own organisation, the one its site or docs link to, and 5 for a checksum on every file.
- 0 to 20, safety evaluation. 20 when results are published and someone other than the publisher ran them or can rerun them, 10 for the publisher's own results, 0 for none.
- 0 to 15, guidance for agent use. 15 for documented prompt-injection guidance or an instruction hierarchy the model was trained to follow, 10 for a statement of what its trained refusals cover and the safeguards to add around it, 5 for general cautions, 0 for none.
- 0 to 15, licence risk. 15 to start, 5 off for each of these, a use policy or field-of-use limits beyond the law, terms the licensor can change or revoke, a revenue or user threshold, and a licence that changed without a dated note or whose link is broken. Not below 0.
- 0 to 10, a security programme. 5 for a way to report a flaw (security.txt, a disclosure policy or a security contact), 5 for advisories handled in public.

When our reference-machine runs start, measured robustness (prompt injection with AgentDojo, over-refusal with XSTest) joins Risk.

## 2. Agent interface, 35 out of 100, up to 10 more on the total

Why it scored 35: Chat template 20, a Jinja file in the repository. Tool-calling format 10, only the inherited template shows the format and no parser is named. Trained tool calling 0, nothing says it was trained to call tools. Structured output 0, not documented. Reasoning control 0, thinking can be switched off but was never trained, and the card asks you to keep it off. Context 5, 40,960 tokens from the base, though Hugging Face trained at 8,192 and under 16K. Most of what the card documents for agents is Carbon-X as a tool another model calls, which this checklist doesn't read.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-interface):

What a model and a developer need to talk to it. Speech, image and other models with no chat interface read the tool-calling and structured-output lines for what the model takes and returns (voices, timestamps, streaming, typed outputs), and the note says how.

- 0 to 20, the chat template. 20 for a template in the repository (Jinja, in the tokenizer config or its own file), 10 when the prompt format is only given as code you run.
- 0 to 25, the tool-calling format. 25 when the publisher documents how tools are declared and how calls come back, with the parser to use in a named runtime, 10 when only the template or a runtime's own recipe says it.
- 0 to 20, tool calling the model was trained for. 20 with several calls in one turn, 10 for one call at a time.
- 0 to 15, structured output. 15 when JSON or JSON Schema output is documented for the model with how to ask for it, 5 when the card claims it without saying how.
- 10, reasoning control. Thinking that can be turned off or budgeted per request, with where the reasoning text comes back.
- 0 to 10, context. 10 for 128K tokens or more trained, 5 for 32K to 127K. Context reached only by scaling the position encoding counts one band down.

## 3. Reach, 90 out of 100, up to 3.8 more on the total

Why it scored 90: Memory 30, the publisher's Q4_K_M GGUF is 9.6 GB, or 11.6 GB with a fifth added. Runtimes 20, released llama.cpp (from b11742, which added the DNA tokenizer), vLLM, SGLang and Transformers all run it, with no Ollama or LM Studio entry. Builds 15, Hugging Face publishes GGUFs at four precisions. Terms 15, Apache 2.0 on the model and on its Qwen3-14B base. Download 10, nothing is gated. Hosted 0, no inference provider or Hugging Face API serves it.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reach):

Whether an agent can run it, and where.

- 0 to 30, memory. The smallest build from the publisher or a runtime's own library that keeps 4-bit precision or better, at its file size plus a fifth for the runtime and a working context. 30 for 16 GB or less, 25 for 24 GB, 20 for 32 GB, 15 for 64 GB, 10 for what fits our reference machine (the GPU share of 128 GB of unified memory), 5 for 256 GB, 0 above.
- 0 to 25, runtimes. 5 for each of llama.cpp, Ollama, vLLM, SGLang, LM Studio and Hugging Face Transformers that runs the architecture in a released version, from the runtime's own code, library or documentation. A publisher's fork or an unreleased branch doesn't count. 25 at most.
- 0 to 15, builds from the publisher. 15 for quantised builds at two precisions or more (FP8, 4-bit, GGUF), 10 for one, 0 when only other people's quantisations exist.
- 0 to 15, terms of use. 15 for an OSI-approved licence with nothing added, 10 for a custom licence that lets anyone use it commercially for free below a threshold or under a use policy, 5 for research or non-commercial use only, 0 for no licence or unclear terms.
- 0 to 10, download. 10 when the files download with no account and no terms to accept, 5 for a click-through that needs an account, 0 when a person approves each request.
- 5, hosted too. The publisher's own API serves it, or at least two inference providers do, so an agent without the hardware can still call it.

When our reference-machine runs start, speed on that machine (tokens per second and time to first token at the build that fits) joins Reach.

## 4. Documentation and upkeep, 77 out of 100, up to 3.5 more on the total

Why it scored 77: Model card 17, everything but a knowledge cut-off. Training data 15, the mid-training sources and shares are listed, the instruction and fine-tuning data aren't named, and there's no cut-off. Last change 20, 22 October 2026. Responsiveness 20, the one question so far got a same-day answer from the organisation. Version policy 5, a first release whose history was squashed without a note. File health 0, the card says llama.cpp isn't supported while the publisher's GGUF repository and launch post say it is, and the tokenizer config's 131,072 contradicts the card's 40,960.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-upkeep):

- 0 to 20, the model card. Intended use, limitations, architecture, parameters, context length, languages and knowledge cut-off. 20 for all seven, 3 off for each that's missing.
- 0 to 20, training data. 20 for the data published or listed with sources and sizes, 15 for a data card that names the sources or the mix, the cut-off and the filtering, 10 for a general description with a cut-off, 5 without one, 0 for none.
- 0 to 20, the last change to the files, the template or the reference code. 20 within 30 days, 15 within 90, 10 within 180, 5 within a year, 0 older.
- 0 to 20, responsiveness. The publisher's replies to the 40 most recent questions and reports in the repository's discussions or issues. 20 for most, 10 for some, 5 for a few, 0 for none.
- 0 to 10, version policy. 10 for earlier versions kept in their own repositories with a dated note when one is replaced, 5 for earlier versions kept without notes or a first release, 0 when files were replaced in place without a note.
- 0 to 10, file health. 10 when config, tokenizer, template and card agree and no report of a broken file waits without a reply, 5 when one does, 0 when the files contradict the card.

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- Capability is pending. The checklist has no DNA or biology area, so the five DNA benchmarks can't count however they're run, and no independent result in any area was found on 23 October 2026; the total is capped at 61
- SGLang is credited from its released qwen3 code. Nobody documents Carbon-X on SGLang, and whether its tokenizer handling runs the custom HybridDNATokenizer correctly wasn't tested
- File health is 0 because the card contradicts the publisher's own GGUF repository on llama.cpp. If Hugging Face fixes the card, it rises to 5 (the model_max_length mismatch remains) or 10
- unchecked, whether Hugging Face handles advisories in public for model repositories, since github.com's advisory pages answer our session with 403
- The main repository's history was squashed into one commit on 22 October 2026, so whether the weight files changed after the 21 October upload couldn't be checked from the history

## Weaknesses

- Not clinically validated and no wet-lab validation reported
- The Python tokenizer needs trust_remote_code, and llama.cpp before b11742 tokenizes DNA wrongly
- Not trained for tool calling, structured output or reasoning
- The card says llama.cpp is unsupported while the publisher's GGUF repository says it is
- No inference provider or hosted API, and no knowledge cut-off

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- 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
- 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
- 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
- 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
- Only A, C, G and T form 6-mers, so filter N and other IUPAC codes first or they become a single oov token
- Strip the dna tags from answers if you need bare sequence, llama.cpp keeps them in the text
- Cap max_tokens for next-base questions, long DNA generation is slow and degrades

## What the review panel asked for

- publish the dual-use evaluation
- a tokenizer without remote code
- correct the llama.cpp line
- note history rewrites
- a JSON output format for answers
- a corrected llama.cpp line on the card

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.