Ornith-1.5-397B

# Fix list: Ornith-1.5-397B

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/ornith-1-5-397b, the October 2026 research run, assessed 21 October 2026. Grade C, 59 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 Ornith-1.5-397B: 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, 50 out of 100, up to 15.4 more on the total

Why it scored 50: File safety 25, safetensors and GGUF only and no Python in any of Ornith's repositories for this size, so the card's --trust-remote-code flag loads nothing from the repository. Provenance 15, the files come from the ornith-ai organisation Ornith's blog links to, with a SHA-256 for every weight file. Safety evaluation 0 and agent guidance 0, nothing published on either. Licence risk 10, MIT carries no policy, threshold or changeable terms, but the card's licence link is broken. Security programme 0, no security.txt, disclosure policy or security 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, 65 out of 100, up to 13.5 more on the total

Why it scored 65: Memory 0, since the smallest 4-bit build, Ollama's at 242 GB, needs about 290 GB with the fifth added, above the 256 GB band, and Ornith's own Q4_K_M GGUF and NVFP4 builds need about the same. Runtimes 25, the most the line gives, from vLLM 0.19.1, SGLang 0.5.9, Transformers 5.8.1, llama.cpp's qwen35moe architecture and Ollama's library, while LM Studio's entry is the base's. Builds 15, Ornith publishes FP8, NVFP4 and GGUF at several precisions. Terms 15, MIT as named in the card's metadata on 21 October 2026, with nothing added, read by its terms though the repository has no licence file. Download 10, nothing is gated. Hosted 0, Ornith runs no API and no inference provider 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, 36 out of 100, up to 9.8 more on the total

Why it scored 36: Model card 11, it covers intended use, architecture, parameters and context but not limitations, languages or a knowledge cut-off. Training data 5, a description of the method with no data or cut-off. Last change 20, Ornith uploaded new shards to the NVFP4 build on 20 October 2026. Responsiveness 0, nobody answered either of the 2 questions. Version policy 0, every GGUF file and the NVFP4 shards were replaced in place in August with no note. File health 0, the card links a missing LICENSE, the two chat templates differ, and the NVFP4 build carries shards its index doesn't use.

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, 85 out of 100, up to 2.3 more on the total

Why it scored 85: Chat template 20, a Jinja file in the repository. Tool-calling format 25, the card shows how tools go in and how calls come back and names the parser for vLLM and SGLang. Trained tool calling 20, Ornith trains it on agent tasks and the template renders several calls in one turn. Structured output 0, not documented or claimed. Reasoning control 10, Ornith's template turns thinking off per request with enable_thinking set to false and the card says the reasoning comes back in reasoning_content, though the card doesn't name the switch. Context 10, 262,144 tokens native.

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 NVFP4 build's eight shards uploaded on 20 October 2026 aren't in its index. They may be the start of a new build that hasn't been finished, which is why they earn last change 20 but cost file health. If an unfinished upload shouldn't count as a change, last change falls to 15 (the 29 August NVFP4 replacement, 53 days) and Documentation and upkeep to 31
- The card's vLLM command passes --trust-remote-code, which loads nothing here because the repository holds no code. If the flag alone should cost the 10 in file safety, Risk falls to 40
- Reasoning control was given 10 because Ornith's own template has the per-request switch and the card says where the reasoning comes back, though the card doesn't name the switch, the same reading as Ornith-1.5-9B. A reading that needs the card to say it gives 0 and Agent interface 75
- No LICENSE file and no notice for Qwen3.5 or Gemma 4 (Apache 2.0). Whether MIT can cover the whole derivative wasn't established
- Too big for the 128 GB reference machine at any precision, so the first batch's note applies: our own Capability runs, speed and measured robustness may stay pending for this listing
- Unchecked, a GitHub organisation or security contact for Ornith, since the GitHub API wasn't available to this session
- Only Hugging Face's provider list and OpenRouter were checked for hosting. Another provider's own list could carry it and earn hosted 5
- Confidence is low by the checklist's rule, since the Capability script found no independent results for this model. Ornith's card compares it with Claude Opus 4.8 on its own runs, which don't count

## Weaknesses

- Needs about 290 GB of memory at 4 bits and about 970 GB in BF16, so it doesn't fit the reference machine
- No LICENSE file though the card links one, and no notice for the Apache 2.0 Qwen3.5 and Gemma 4 work it builds on
- No safety evaluation, agent-use guidance or security contact
- Every GGUF and the NVFP4 shards were replaced in place in August 2026 with no note, and the NVFP4 build now holds eight shards its index doesn't reference
- The two chat templates disagree on keeping earlier reasoning, structured output isn't documented and the card doesn't give the active parameters

## 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.

- Start it with the card's vLLM command at `--tensor-parallel-size 8`, or `ollama run ornith-1.5:397b` on a machine with about 300 GB of memory
- In vLLM pass `--enable-auto-tool-choice --tool-call-parser qwen3_xml --reasoning-parser qwen3`, and in SGLang `--tool-call-parser qwen3_coder --reasoning-parser qwen3`, so tool calls and reasoning come back parsed
- Thinking is on by default. Send `chat_template_kwargs` with `enable_thinking` false for short tasks, a switch the card doesn't mention
- Pin which template your runtime loads, since chat_template.jinja keeps earlier reasoning and the tokenizer_config.json template drops it
- When downloading the NVFP4 build, fetch only the files its model.safetensors.index.json names (the 25-shard set, 241.7 GB), not the eight extra shards
- Pin a revision when you download, since Ornith has replaced files in place
- Validate JSON output yourself, since structured output isn't documented

## What the review panel asked for

- finish or remove the new shards
- date each replacement on the card
- a hosted endpoint
- a LICENSE file
- state the active parameters
- finish or remove the NVFP4 upload
- 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.