Darwin-36B-Opus

# Fix list: Darwin-36B-Opus

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/darwin-36b-opus, the October 2026 research run, assessed 21 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 Darwin-36B-Opus: 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, 55 out of 100, up to 13.8 more on the total

Why it scored 55: File safety 25, safetensors and GGUF only with no code in the repository, so the card's trust_remote_code flag loads nothing extra. Provenance 15, the files are in the FINAL-Bench organisation that vidraft.net links to, with a SHA-256 for every weight file. Safety evaluation 0 and agent guidance 0, nothing published. Licence risk 15, Apache 2.0 through the whole chain with no conditions added. Security programme 0, no security.txt, policy or contact.

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. Reach, 80 out of 100, up to 7.7 more on the total

Why it scored 80: Memory 20, the publisher's Q4_K_M GGUF is 21.2 GB and needs about 25.4 GB with the fifth added, so 32 GB. Runtimes 20, the text-only Qwen3.5 MoE class runs in released vLLM, SGLang, Transformers and llama.cpp, while Ollama's and LM Studio's entries are the base's, not this model's. Builds 15, FINAL-Bench's POCKET-35B-GGUF has it at four precisions. Terms 15, Apache 2.0 with nothing added, read by its terms though there is no licence file. Download 10, nothing is gated. Hosted 0, no provider or 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.

## 3. Documentation and upkeep, 51 out of 100, up to 7.5 more on the total

Why it scored 51: Model card 11, it covers architecture, parameters, context and languages but not intended use, limitations or a knowledge cut-off. Training data 5, the reasoning parent's fine-tune is described without datasets or a cut-off. Last change 10, the publisher's GGUF build went up on 22 July 2026, 91 days ago, and the main weights date from 23 April. Responsiveness 20, the publisher answered 8 of the 13 discussions, and two of the rest are its own posts. Version policy 5, a first release whose weights were replaced twice in its first day with commit titles but no card note. File health 0, the card's head counts, base and evaluation method disagree with the files, and the generation config drops the base's end-of-turn stop token.

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.

## 4. Agent interface, 70 out of 100, up to 4.6 more on the total

Why it scored 70: Chat template 20, Qwen3.6's Jinja template in the repository. Tool-calling format 10, the format is only in the template and the card says nothing about tools. Trained tool calling 20, inherited from the base that was trained for it, with several calls in one turn in the template, though the publisher neither claims nor tests it. Structured output 0. Reasoning control 10, the template turns thinking off per request and the card shows where the reasoning comes back. Context 10, 262,144 tokens.

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.

## 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.

- The commit that put the current weights in (23 April 2026) says Darwin V7 merge plus targeted refinement, and the card says the merge needs no gradient training. Whether any training was done after the merge isn't stated
- Last change is read from the POCKET-35B-GGUF upload (91 days, 10). Counting the card edit of 25 July (88 days) would give 15
- Version policy is read as a first release (5), since the in-place replacements happened in the first 26 hours and the commit titles name them. A stricter reading of files replaced in place without a card note gives 0
- Trained tool calling is credited from the base's training (20), since the publisher claims none. No independent tool-calling result for the merged model was found
- The generation config's single stop token wasn't tested, since nothing was run

## Weaknesses

- The merge method is proprietary and the current weights include an undescribed targeted refinement
- No safety results, agent guidance or security contact
- The card's head counts, base and evaluation method disagree with the files
- Tool calling and structured output aren't documented, and the card names no parser
- No LICENSE file, and the files were replaced in place on the first day with no note on the card

## 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 FINAL-Bench/POCKET-35B-GGUF:Q4_K_M --jinja`, about 21 GB of files
- On the reference machine BF16 fits, `vllm serve FINAL-Bench/Darwin-36B-Opus --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder`, the parsers for Qwen3.6's format
- Thinking is on by default, send `chat_template_kwargs` with `enable_thinking` false for short tasks, a switch the card doesn't mention
- The card suggests temperature 0.6 to 0.7 and at least 5,120 new tokens for its think traces
- generation_config.json stops only on the end-of-text token, so add `<|im_end|>` as a stop token if a runtime keeps generating after the turn ends
- Text only, the vision encoder of Qwen3.6-35B-A3B isn't in these files
- Validate JSON output yourself, since structured output isn't documented

## What the review panel asked for

- correct the card
- add the end-of-turn stop
- a LICENSE file in the repository
- a tool-calling test of the merge
- describe the refinement step
- a security contact

## 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.