gpt-oss-120b

# Fix list: gpt-oss-120b

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/gpt-oss-120b, the October 2026 research run, assessed 21 October 2026. Grade BB, 70.2 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 gpt-oss-120b: 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. Capability, 54 out of 100, up to 16.1 more on the total

Why it scored 54: From 22 independent benchmark results in 5 areas, each scored as its share of the way from the benchmark's floor to the best result any model has on the same leaderboard. By area, Instruction following 83, Reasoning and knowledge 68, Coding 59, Long context 59, Tool calling and agent tasks 30. Publisher-submitted results count at half weight, and numbers that appear only on the publisher's card don't count.

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

How well the model does on the benchmarks that matter to an agent. We don't average a publisher's own numbers into a grade. A result counts in full when we ran it on our reference machine, or when an independent party ran it and published the method (the Berkeley Function Calling Leaderboard, Artificial Analysis, Epoch AI's Benchmarking Hub, the SWE-bench and Terminal-Bench leaderboards, the Open ASR Leaderboard and others that run models themselves). A result the publisher submitted to such a leaderboard counts at half weight. A number that appears only on the publisher's card is shown on the listing as the publisher's claim and doesn't count.

Each result is scored as its share of the way from the benchmark's floor (chance for multiple choice, zero for a pass rate) to the best result any model, open or closed, has on the same leaderboard and setting on the day we read it. A model that matches the best model anywhere scores 100 on that benchmark. For an error rate such as word error rate, lower is better and the scale runs the other way, and where the leaderboard gives no floor the rate is read against the best model's, so twice the best model's error rate scores 50. Speed as a multiple of real time (RTFx) runs from about 1 to over 20,000, so it's read on a log scale from real time to the fastest model. On the MTEB leaderboard, where most models score between 0.5 and 0.75, the floor is the median of every model the same tab lists, and its results count at half weight because MTEB doesn't say who ran them and most are submitted with the model. An arena rating (Elo) isn't a ratio, so its floor is 400 points below the best model on that arena, the gap at which the better model wins about ten games in eleven. Where a model has results at several settings (reasoning effort, say), its best one counts, because that's the setting an agent would run.

The results group into areas, each weighted within the category.

- Tool calling and agent tasks, 35 (BFCL, τ²-bench, Terminal-Bench and agent suites like them)
- Coding, 20 (SWE-bench Verified, LiveCodeBench, Aider Polyglot)
- Reasoning and knowledge, 20 (GPQA Diamond, MMLU-Pro, Humanity's Last Exam, composite indexes that publish their components)
- Instruction following, 15 (IFEval, IFBench)
- Long context, 10 (RULER, LongBench and long-context retrieval tests)

A speech model is read on its own areas, recognition accuracy (word error rate on the Open ASR Leaderboard's sets, 60) and speed (real-time factor, 40) for speech to text, and intelligibility and speaker similarity (Seed-TTS-eval, the Open TTS Leaderboard, 70) and speed (30) for text to speech. A publisher's own converted copy of the same weights (Qwen's -hf builds, say) counts as the same checkpoint. An embedding model is read on retrieval (60) and the other MTEB task types, classification, clustering, reranking and similarity among them (40), from the MTEB leaderboard. A vision model adds an image area (MMMU, ChartQA, DocVQA or OCRBench) at 20, and the others shrink to fit.

Capability is the weighted mean of the areas with results, and needs results in at least two areas (one for a speech or embedding model). With less than that it's pending for that listing, its weight is shared across the other four, and the total is capped at 61, so a model nobody has benchmarked independently can't reach B, and every model with results ranked B or better stays above it. The listing says so. Until 21 October 2026 the cap was 69, which put models nobody had benchmarked above ones that had been.

## 2. Reach, 75 out of 100, up to 6.3 more on the total

Why it scored 75: Memory earns 10, because the smallest 4-bit build, Ollama's 65 GB gpt-oss:120b, needs about 78 GB with the fifth added, which fits the reference machine but not 64 GB. Runtimes earn the full 25 from Transformers, vLLM, SGLang, llama.cpp and Ollama. Builds earn 10, since OpenAI ships one precision, MXFP4. Terms earn 15 for Apache 2.0 with a policy that asks only for lawful use. Download earns 10, with no gate or account. Hosted earns 5, since 11 inference providers serve it on Hugging Face even though OpenAI's own API doesn't.

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, 45 out of 100, up to 5.5 more on the total

Why it scored 45: The model card earns 20, with all seven items in OpenAI's arXiv card. Training data earns 10, a general description with a cut-off. Last change earns 5, since the last change to code that runs the model was 1 November 2025 and the files date from August 2025. Responsiveness earns 0, with no OpenAI reply in the 40 latest discussions. Version policy earns 5 as a first release. File health earns 5, the files agree but template and tokenizer reports wait without a reply.

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. Risk, 95 out of 100, up to 1 more on the total

Why it scored 95: File safety earns 25, safetensors only, with metal/model.bin a raw format and not pickle, and nothing from the repository runs on load. Provenance earns 15, files in OpenAI's linked organisation with a hash for every file. Safety evaluation earns 20, published results with SecureBio as an outside evaluator. Agent guidance earns 15 for the trained instruction hierarchy with prompt-injection results. Licence risk keeps all 15. The security programme earns 5 for OpenAI's Bugcrowd programme and disclosure policy, and the advisories half scores 0 because GitHub couldn't be checked, not because it was found missing.

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.

## 5. Agent interface, 90 out of 100, up to 1 more on the total

Why it scored 90: The chat template earns 20, a Jinja file in the repository. The tool-calling format earns 25, since OpenAI documents declarations and returned calls in harmony and tells vLLM users to parse with openai-harmony. Trained tool calling earns 10, because each call stops generation, so calls come one at a time. Structured output earns 15 for the documented Response Formats section with a JSON Schema. Reasoning control earns 10, three effort levels per request with the reasoning on the analysis channel. Context earns 10 for 131,072 trained 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.

- unchecked, public security advisories for openai/gpt-oss, since GitHub's pages answered 403 to this session, so that half of the security programme scores 0
- unchecked, replies to GitHub issues on openai/gpt-oss, so responsiveness rests on the Hugging Face discussions only
- Terms read the USAGE_POLICY as adding nothing beyond the law, as the calibration pass did. If any policy file counts as added, terms drop to 10 and Reach to 70
- Reasoning control reads three effort levels as a budget per request. If only an off switch counts, it's 0 and Agent interface drops to 80
- The Transformers route fetches kernels-community/gpt-oss-triton-kernels from the Hub to run MXFP4, code from outside the model repository, so no deduction was taken, and gpt-oss-20b is read the same way. A reviewer may read it otherwise, which would take 10 off file safety for both
- hardware.json gives 96 GB as the memory estimate for both builds, while this grading uses the brief's size × 1.2, 78 GB. Both sit in the reference-machine band

## Weaknesses

- Needs about 78 GB, too big for 64 GB machines, and OpenAI ships only one precision
- One tool call at a time
- No OpenAI reply in the last 40 discussions, including open template and tokenizer reports
- No change to the files since August 2025 or to the model code since November 2025
- Reasoning can be lowered but not turned 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.

- Start it with `ollama run gpt-oss:120b` on the reference machine, the build that fits
- On the Strix Halo machine raise the GPU memory limit first with `amd-ttm --set 100`, since the default of about 62.8 GiB is below the 78 GB it needs
- With vLLM use `vllm serve openai/gpt-oss-120b --tool-call-parser openai --enable-auto-tool-choice` per vLLM's recipe, or parse raw output with openai-harmony
- Always use the harmony format through the chat template or openai-harmony, the card says it won't work correctly otherwise
- Set `Reasoning: low` in the system message for short tasks, medium is the default
- Pass the reasoning back after a tool call until the final answer, and keep the analysis channel away from end users
- Expect one tool call per turn and loop, don't wait for parallel calls

## What the review panel asked for

- answer the tokenizer reports
- parallel tool calls in harmony
- ship the MXFP4 kernels in the model repository

## 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.