SmolLM3-3B

# Fix list: SmolLM3-3B

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/smollm3-3b, 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 SmolLM3-3B: 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 25, safetensors only and nothing from the repository runs on load or to build prompts. Provenance 15, the HuggingFaceTB organisation linked from Hugging Face's launch post, with a SHA-256 for each shard. Safety evaluation 0, none published. Agent guidance 5, only the general cautions in the card's Limitations section. Licence risk 15, Apache 2.0 with nothing added and a working licence link. Security programme 5, huggingface.co has a valid security.txt and Transformers has a security policy; whether advisories are handled in public couldn't be checked from our session, so that half is scored as absent.

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. Documentation and upkeep, 54 out of 100, up to 7.1 more on the total

Why it scored 54: Model card 14, it covers limitations, architecture, parameters, context and languages but has no intended-use section and leaves the June 2025 cut-off to the chat template. Training data 20, the datasets, mixture weights and post-training set are published with sources and sizes. Last change 0, the files last changed on 10 September 2025, more than a year ago, and the reference code has had only library-wide refactors since. Responsiveness 10, staff answered 14 of 35 discussions, but only one since October 2025. Version policy 5, a first release never replaced, with no dated notes. File health 5, template and tokenizer reports wait with no 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.

## 3. Reach, 85 out of 100, up to 5.8 more on the total

Why it scored 85: Memory 30, the smallest 4-bit build Hugging Face publishes is its own ONNX q4f16 at 2.1 GB, about 2.5 GB with the fifth added, and the full BF16 files that the counted runtimes load come to about 7.4 GB, both under 16 GB. Runtimes 20, released Transformers carries the model code, llama.cpp knows the architecture, and vLLM and SGLang run it through their Transformers backends in released versions; Ollama has no library entry and LM Studio's catalogue has none, so those two earn nothing. Builds 10, Hugging Face's ONNX repository has 4-bit builds, one of the checklist's kinds, and int8 builds, which are neither FP8 nor GGUF. They run in ONNX Runtime and Transformers.js rather than the runtimes above. Terms 15, plain Apache 2.0, scored by its terms since the repository holds no LICENSE file. Download 10, not gated. Hosted 0, no inference provider is listed for it on the Hub and none of the provider lists checked carries 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. Agent interface, 80 out of 100, up to 3.1 more on the total

Why it scored 80: Chat template 20, a Jinja file in the repository. Tool-calling format 25, the card says how tools are declared, how calls come back and which vLLM parser to use. Trained tool calling 20, the template invites one or more calls per turn, each in its own block. Structured output 0, nothing documents asking the weights for JSON. Reasoning control 10, thinking turns off per request through the kwarg or a system-prompt flag, and the reasoning comes back in think tags. Context 5, trained at 65,536 tokens, with 131,072 only through YaRN.

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, whether Hugging Face handles Transformers advisories in public in its own repository; GitHub isn't reachable from our session, so that half of security programme is scored as absent
- Unchecked, whether Transformers' smollm3 code had a model-specific change since 21 October 2025; the research record calls the recent changes library-wide refactors, and GitHub couldn't be re-read here, so last change rests on the repository's own files
- Ollama's pinned llama.cpp knows the smollm3 architecture, so a GGUF probably loads through hf.co, but with no library entry the line scores 0
- Responsiveness rests on the Hugging Face discussions only; issues on github.com/huggingface/smollm couldn't be read
- The card names six main languages including German, while its metadata leaves German out and lists Chinese, Arabic and Russian
- Confidence follows Capability's independent areas, which the script computes; medium is carried from the earlier reading and may change
- Builds read the ONNX repository by the checklist's kinds, 4-bit only, so 10. If a build must run in one of the counted runtimes to count, it's 0, and if int8 counts as a second precision, 15

## Weaknesses

- 65,536 tokens trained; 131,072 needs a YaRN edit to config.json
- No change to the files since 10 September 2025 and no publisher reply in discussions since November 2025
- No safety evaluation or agent-use guidance
- No structured-output documentation
- No Ollama or LM Studio catalogue entry and no hosted provider
- The chat template renders tool results as user turns and drops tool_calls from earlier assistant messages

## 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 `vllm serve HuggingFaceTB/SmolLM3-3B --enable-auto-tool-choice --tool-call-parser=hermes`, or tool calls come back as text
- Pass `chat_template_kwargs` with `enable_thinking` false for short tasks, since thinking is on by default; a `/think` or `/no_think` flag in the system prompt overrides the kwarg
- Put earlier tool calls in the assistant message content, since the template ignores a `tool_calls` field when it renders history
- Add the YaRN `rope_scaling` block (factor 2.0, original 65536) to config.json before sending more than 65,536 tokens
- Use temperature 0.6 and top_p 0.95, the card's recommended sampling
- Needs Transformers v4.53.0 or later; for llama.cpp use a GGUF from ggml-org or another converter, since Hugging Face publishes none
- Validate JSON output yourself or use the runtime's guided decoding

## What the review panel asked for

- answer the template reports
- a GGUF build from Hugging Face
- a knowledge cut-off on the card
- safety results 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.