# Fix list: Gemma 4 31B IT From Anchor Terminal's listing at https://www.anchorterminal.com/tools/gemma-4-31b-it, the October 2026 research run, assessed 21 October 2026. Grade BB, 74.1 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 Gemma 4 31B IT: 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, 61 out of 100, up to 13.7 more on the total Why it scored 61: From 14 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 91, Long context 79, Coding 68, Reasoning and knowledge 51, Tool calling and agent tasks 45. 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. Risk, 70 out of 100, up to 6 more on the total Why it scored 70: File safety 25, safetensors only and no code from the repository runs to load it or build prompts. Provenance 15, the files come from the google organisation that gemma.ai links to, with a SHA-256 for every weight file and a git hash for every file. Safety evaluation 0, Google says violations were minimal but publishes no results. Agent guidance 10, the card says what its safety training covers and asks for content safeguards, and the tool guide says to validate calls, but there is no prompt-injection guidance. Licence risk 15, Apache 2.0 carries none of the deductions and its link works. Security programme 5, Google's company-wide security.txt and reward programme exist but nothing ties them to Gemma, and no public advisories were found. 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. ## 3. Documentation and upkeep, 65 out of 100, up to 3.5 more on the total Why it scored 65: Model card 20, all seven items are there, with languages given as counts. Training data 10, a general description with a January 2025 cut-off. Last change 10, the files last changed on 20 July and Google's reference code for Gemma 4 last changed in a way that runs this model on 6 July, both inside 180 days. Responsiveness 10, Google answered 14 of the last 40 discussions, none since 30 July. Version policy 10, earlier generations are kept in their own repositories with dated releases. File health 5, two template 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. Agent interface, 85 out of 100, up to 1.5 more on the total Why it scored 85: Chat template 20, a Jinja file in the repository. Tool-calling format 25, Google documents the declaration and call syntax and ships a response schema that Transformers parses. Trained tool calling 20, several calls and responses in one turn. Structured output 0, Google documents none and the card doesn't claim it. Reasoning control 10, thinking turns on or off per request and comes back in its own block. 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. ## 5. Reach, 95 out of 100, up to 1.3 more on the total Why it scored 95: Memory 25, Google's own QAT Q4_0 GGUF is 17.7 GB, which with a fifth added comes to about 21 GB and fits a 24 GB card even with the vision projector. Runtimes 25, released code in llama.cpp, Ollama's library, vLLM, SGLang and Transformers fills the cap without LM Studio, which only Google's guides name. Builds 15, Google ships a 4-bit GGUF and a 4-bit compressed-tensors build, read as two of the kinds the checklist names. Terms 15, plain Apache 2.0 with nothing added. Download 10, no gate and no account. Hosted 5, Google's Gemini API serves it today. 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. ## 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 Google's vulnerability reward programme takes reports about Gemma's open weights, since its rules page renders only with JavaScript - unchecked, github.com/google-gemma linked from the card answered 403 to our reader, so any security policy or advisories there weren't read - builds read as two kinds because the checklist lists GGUF apart from 4-bit, though both of Google's builds are 4-bit. A stricter reading gives 10 - the two template reports from 29 April and 10 June may be covered by the 15 July template fix, which would lift file health to 10 - confidence follows Capability's independent areas, which the script computes, so medium is carried from the old grading ## Weaknesses - No safety evaluation results published - No prompt-injection guidance and no security route tied to Gemma - Structured output isn't documented by Google - Training data described only in general terms - Files unchanged since 20 July and no Google reply in the discussions since 30 July ## 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. - On the reference machine run `llama-server -hf google/gemma-4-31B-it-qat-q4_0-gguf --hf-file gemma-4-31B_q4_0-it.gguf --jinja -c 32768`, adding gemma-4-31B-it-mmproj.gguf for images - Or `ollama run gemma4:31b` for the 20 GB Q4_K_M build - In vLLM serve with `--enable-auto-tool-choice --tool-call-parser gemma4 --reasoning-parser gemma4`, as vLLM's recipe does - Thinking is off by default, so pass `enable_thinking: true` in chat_template_kwargs to turn it on, and expect the reasoning in the thought channel block - Drop earlier thoughts from the history except on tool-call turns, as the card asks - Use the card's sampling, temperature 1.0, top_p 0.95, top_k 64 - Validate function names and arguments before running a call, and check JSON output yourself or use the runtime's guided decoding - The w4a16 build relies on Marlin INT4 kernels that vLLM marks unsupported on AMD, so use the GGUF on the reference machine ## What the review panel asked for - update the card's links - close the template reports - KV cache figures in the memory table - update the card's links to gemma.ai - link a security contact from 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.