Iris-3B

# Fix list: Iris-3B

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/iris-3b, the October 2026 research run, assessed 15 October 2026. Grade D, 50.9 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 Iris-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. Reliability, 45 out of 100, up to 11 more on the total

Why it scored 45: Read as local software, since Iris-3B is a model the owner runs. No package on PyPI. Install is a git clone and `pip install -e .`, with Python 3.11 or newer, PyTorch 2.7.1 or newer and an NVIDIA GPU with CUDA stated (12 of 20). The repository has 23 pytest files covering the model, flow, checkpoints and parallel training, but no CI workflow, so nothing shows them passing on main (5 of 25). Three GitHub issues in ten days, none a crash, two closed and one open announcing a community ComfyUI port. Six Hugging Face discussions, none a bug report (20 of 25). No tags, releases or changelog. Package version 0.1.0, and new checkpoints are added in folders of one Hub repository with no version pins (3 of 15). Released as a final checkpoint with no beta label, but the package is 0.1.0 and the project ten days old (5 of 15).

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

Hosted APIs, MCP servers, models and platforms.

- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).
- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.
- 15, rate limits documented with numbers.
- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.
- 10, an SLA published for any paid tier.
- 10, the surface agents use is generally available, not beta or preview.

Local packages, SDKs, frameworks and stdio MCP servers.

- 20, installs from an official package with supported runtimes stated.
- 25, a public CI and test suite, passing on the default branch.
- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).
- 15, semver discipline and breaking changes called out in a changelog.
- 15, version 1.0 or later, or declared stable.

Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.

## 2. Security & auth, 44 out of 100, up to 9.8 more on the total

Why it scored 44: Read as software the owner runs. No account, key or token is needed to get or run the weights, and the software opens no network service (14 of 30). No write actions beyond image files in the output folder (15 of 20). The output is an image, not text an agent reads as instructions. The card warns of inaccurate, biased or unsafe output, and neither the code nor the demo has a content filter (8 of 15). No call log. Output files are named after the prompt (3 of 15). No SECURITY.md, security.txt, bug bounty or advisories. Released checkpoints are safetensors. Training `.pth` files, which the CLI also accepts, are read with `torch.load(weights_only=False)`. Dion is pinned to a commit (4 of 20).

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

- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.
- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.
- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.
- 0 to 15, audit logs or per-call visibility for the operator.
- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.

Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing.

## 3. Schema & documentation, 47 out of 100, up to 8.6 more on the total

Why it scored 47: Read for a model the owner runs through a CLI and a Python package. No server, OpenAPI or other machine-readable contract. The CLI uses argparse with typed flags and help text, and the Python functions carry type hints. The demo Spaces' Gradio API lists typed endpoints, but Speridlabs doesn't document it (8 of 25). No llms.txt. The README and model card are Markdown in each repository (5 of 10). The card and README state limits (unreliable rendered text and object counts, about one megapixel, depth relative not metric, Turbo's fixed schedule and lower seed variety) but not intended or out-of-scope uses (14 of 20). Flags are typed, with the size presets as an enum only in the Spaces. Height and width take any integer, though Turbo was distilled at seven sizes (9 of 15). Commands for every model and a tips table, but no documented errors (7 of 15). No versioning or changelog beyond dated commits (4 of 15).

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

APIs and MCP servers.

- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).
- 10, llms.txt or Markdown docs served for agents.
- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.
- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.
- 0 to 15, examples and documented error responses.
- 15, versioning and a public changelog.

Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference.

## 4. Agent ergonomics, 54 out of 100, up to 7.5 more on the total

Why it scored 54: Read as a local image model an agent calls through the CLI or Python. Images are written to files with a contact sheet, so nothing large returns into context. Batch prompts through `--txt-file` (15 of 25). Output size through `--height` and `--width`, an output folder, `--max-side` for depth and a tile budget for restoration with `--no-budget` to lift it (12 of 20). No documented error responses. A Turbo checkpoint with incompatible flags exits with a message naming the flags to drop (6 of 20). A fixed `--seed` reproduces an image and there are no side effects beyond files, but files are named after the prompt and overwritten on repeat (14 of 20). Sensible defaults and two required inputs (`--checkpoint` and a prompt). No SDK beyond the Python package, which isn't on PyPI, and about 12 GB of weights before a first image (7 of 15).

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

- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).
- 20, pagination, filtering and output-size controls.
- 20, actionable, documented error responses, codes and messages an agent can recover from.
- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.
- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.

Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs.

## 5. Payments & pricing, 60 out of 100, up to 5 more on the total

Why it scored 60: Free Apache-2.0 software with nothing to buy, so 20 + 20 + 20 for pricing, free use and no sign-up. No payment protocol (0). There is no hosted API. Hardware is the owner's cost.

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

The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).

- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.
- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login.
- 20, a free tier or trial that doesn't need a card.
- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).

Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.

Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol.

## 6. Transparency & trust, 47 out of 100, up to 4.6 more on the total

Made of editorial 50, provenance 44.

Why it scored 47: Apache-2.0 for code and weights, with a NOTICE file, a third-party licence table and the full training and distillation recipe. The training images and the Turbo prompt sets are not released or named in the README or card (25 of 30). Self-hosted, so inputs stay on the owner's hardware, but we found no statement saying so and speridlabs.com has no privacy policy (15 of 30). No deprecation policy, version tags or pinned revisions (2 of 20). No telemetry calls found in the inference code. Training defaults to `report_to: wandb`, which sends run config and validation images to Weights & Biases, can be set to `none`, and is not mentioned in the README (8 of 20).

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

- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.
- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).
- 0 to 20, a deprecation policy or notices with dates.
- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).

The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.

Provenance checks not met in full (half of this category, computed from checked facts):

- Domain age: speridlabs.com, registered 2025-02-25 (1 year) (3 of 15)
- Status page: not found (0 of 10)
- Changelog: not found (0 of 10)
- security.txt: not found (0 of 10)

## 7. Maintenance & community, 70 out of 100, up to 2.6 more on the total

Why it scored 70: Read for an open-weight model ten days old. Turbo weights on 12 October and a demo fix on 15 October 2026 (30). Three dated weight releases, base on 5 October, depth and restoration on 8 October and Turbo on 12 October (20). The project lead answered and closed issue 1 within a day, and two outside pull requests merged within two days. Issue 3 and the three Hugging Face discussions we read had no reply from Speridlabs (14 of 25). No SDK and no PyPI package. Community ComfyUI packs and quantised conversions exist but are not the vendor's (2 of 15). Dependencies are current (PyTorch 2.7.1, transformers 4.57), but there's no CI and `wandb` is a required dependency for inference (4 of 10).

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

- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.
- 20, at least three releases or dated changelog entries in the last 90 days.
- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.
- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).
- 10, package health, current dependencies and CI.

Models are read for deprecation notice periods and model churn rather than release counts.

## 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: memory needed for text-to-image alone. The demo README gives 45 GB for all three models and 20 GB peak with offloading.
- unchecked: whether Speridlabs replied to Hugging Face discussions 2, 3 and 5. We read 1, 4 and 6.
- unchecked: today's star and fork counts beyond the GitHub API listing read early on 15 October, because the API then rate-limited us.
- The training data is not named in the README or the model card. We didn't read the paper's data section in full.
- Whether the Gradio demo Spaces count as an interface for agents. They expose a typed Gradio API on Hugging Face ZeroGPU, but Speridlabs doesn't document it as one.

## Weaknesses

- No package on PyPI, no version tags, no changelog and no CI workflow. Install is a git clone and `pip install -e .`
- Needs an NVIDIA GPU with CUDA. All three models together take about 45 GB of GPU memory, or about 20 GB peak with offloading
- No HTTP server, OpenAPI or MCP interface. The only hosted route is the Gradio demo Spaces on Hugging Face ZeroGPU
- No safety filter in the code, no SECURITY.md or security.txt, and no privacy or terms pages on speridlabs.com
- Training defaults to reporting runs to Weights & Biases (`report_to: wandb`), which the README doesn't mention

## 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.

- Check for an NVIDIA GPU with CUDA and Python 3.11 or newer before downloading. The base weights are about 12 GB, and Qwen3-VL-4B downloads on the first run
- Exclude `depth/*`, `upscaler/*` and `turbo/*` in `hf download` unless those models are needed. Each of the first two adds about 12 GB
- Don't pass `--steps`, `--cfg-scale`, `--negative-prompt`, `--order` or `--shift` with a Turbo checkpoint. The script exits with an error
- Pass `--seed` to reproduce an image, and set `--outdir`, since files are named after the prompt and a repeat run overwrites them
- Load only exported directories (`model.safetensors` and `config.yaml`). A training `.pth` is read with `torch.load(weights_only=False)`, so never load one from an untrusted source

## 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.