Kolibri-1

# Fix list: Kolibri-1

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/kolibri-1, the October 2026 research run, assessed 20 October 2026. Grade BB, 75.8 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 Kolibri-1: 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. Security & auth, 55 out of 100, up to 7.9 more on the total

Why it scored 55: Read for open weights. File safety 30, safetensors only, no pickle files, and loading needs a pip package rather than code from the repository. Integrity 15, files in Aleph Alpha's own organisation with a SHA-256 for each shard. Safety evaluation 0, the card says refusals were measured in its safety evaluations but publishes no results. Guidance for agent use 10 of 15, the card says the trained safeguards don't replace safeguards in the application and asks the calling system to validate tool results, with no prompt-injection guidance. Security programme 0, no disclosure policy found and security.txt answered 403 to our reader (unchecked).

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.

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

Why it scored 60: Read for open weights, by the self-hosted rule, so 60 is the most a model can score here. No payment protocol (0). Apache 2.0, the full terms public with no login (20). Commercial use free (20). The files download without an account or a gate (20).

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.

## 3. Reliability, 82 out of 100, up to 3.6 more on the total

Why it scored 82: Read for open weights. Runtimes 12 of 30. vLLM through Aleph Alpha's own plugin (6) and llama.cpp, whose master branch knows the kolibri1 architecture (6). No SGLang or mlx-lm model file and no Ollama library entry were found (0 each). A reference implementation in a released Transformers, 5.20.0 (20). Sampling parameters and a serving command published (15). Memory and accelerators stated on the card, about 78 GB at FP8 and the GPU set-ups it needs (15). The vendor's own FP8 build beside BF16 (10). A general release dated 3 October 2026 (10).

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.

## 4. Agent ergonomics, 85 out of 100, up to 2.4 more on the total

Why it scored 85: Read for open weights. Context 262,144 tokens native (25). Tool calling it was trained for, and the template lets it make more than one call per turn (25). Structured output 5 of 15, the card names structured extraction and structured-output capabilities without saying how to ask for them. Reasoning effort none to high per request (15). Fit 15 of 20, the vendor's FP8 build is about 78 GB, within 128 GB, and no vendor or runtime build fits in 32 GB.

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. Transparency & trust, 74 out of 100, up to 2.3 more on the total

Made of editorial 80, provenance 68.

Why it scored 74: Editorial half, read for open weights. Licence 30, Apache 2.0 with no acceptable use policy added. Training data 20 of 30, a data card with the mix for each stage, the curation steps, the synthetic-data models and a cut-off, plus the EU training-content summary, but no list of datasets. Evaluations 20, the eval-framework and Harbor set-ups are published and the same set-up was used for every compared model. Version policy 10 of 20, a first release with nothing replaced yet and no stated policy.

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: aleph-alpha.com, registered 2018-08-11 (8 years) (11 of 15)
- Status page: not found (0 of 10)
- security.txt: could not be fetched (0 of 10)

## 6. Maintenance & community, 75 out of 100, up to 2.2 more on the total

Why it scored 75: Read for open weights. Last change 19 October 2026 (30). At least three dated changes in 90 days that say what they fixed, the release on 3 October, a quantisation config fix on 13 October and transformers_version on 19 October (20). Responsiveness 5 of 25, 2 of the 11 discussions that aren't pull requests had a reply from an Aleph Alpha member by 20 October. Runtime support it maintains, the aleph-alpha-inference plugin 1.0.0 with CI and a container image (15). File health 5 of 10, a report that the tool-call parser in the container image is broken (discussion 8, 7 October) had no reply from Aleph Alpha.

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.

## 7. Schema & documentation, 95 out of 100, up to 0.8 more on the total

Why it scored 95: Read for open weights. The card covers intended use, limitations, architecture, parameters, context, languages and knowledge cut-off (25). A chat template in tokenizer_config.json (20). The tool-calling format documented, Hermes-style JSON in `<tool_call>` tags, with the `kolibri1` parser for vLLM (20). A serving command, a request and a full tool-calling example (15). Reasoning effort and how to turn thinking off documented (10). Versions 5 of 10, a commit history whose messages say what changed but no named versions of the files.

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.

## 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: aleph-alpha.com/.well-known/security.txt answered 403 to our reader, so a security contact couldn't be confirmed
- unchecked: whether the community MLX builds run on a released mlx-lm, since mlx-lm's main branch has no kolibri1 model file
- unchecked: whether vLLM's next release loads Kolibri without the plugin; its main branch maps the architecture to the Transformers backend
- Performance and Task success are pending until our local runs; the FP8 build fits our 128 GB machine

## Weaknesses

- No safety evaluation results published, though the card says refusals were measured
- Runs on vLLM only through Aleph Alpha's plugin, and no Ollama, SGLang or mlx-lm support was found
- 2 of 11 discussions answered by Aleph Alpha, and a report of a broken tool-call parser in the container image had no reply
- Structured output is claimed on the card without saying how to request it
- In Aleph Alpha's own table it trails several compared models on BFCL v4 multi-turn (47.5) and TerminalBench 2.1 (27.7), the vendor's numbers

## 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.

- Serve it with `--tool-call-parser kolibri1 --reasoning-parser kolibri1 --enable-auto-tool-choice`, or tool calls come back as text
- Install aleph-alpha-inference for vLLM 0.29, or use a Transformers release from 5.20.0
- Set `chat_template_kwargs` to `{"enable_thinking": false}` or `reasoning_effort: "none"` for short tasks, since thinking is on by default
- Keep contexts to 262,144 tokens unless you need more, as Aleph Alpha recommends, and pass the config override to go further
- Validate tool results in your own code, since the card says the trained safeguards don't replace application-level checks

## What the review panel asked for

- reply to the parser report
- named file versions
- an official 4-bit build
- a JSON Schema output example
- publish safety evaluation results
- a disclosure policy

## 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.