# Fix list: DeepSeek V4.1 Flash From Anchor Terminal's listing at https://www.anchorterminal.com/tools/deepseek-v4-1-flash, the October 2026 research run, assessed 20 October 2026. Grade B, 62.8 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 DeepSeek V4.1 Flash: 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, 35 out of 100, up to 11.4 more on the total Why it scored 35: Read for open weights. File safety 20 of 30, safetensors only with no pickle files (30), less 10 because the documented way to build prompts is to run DeepSeek's encoding.py from the repository, or its deepseek-recipe library, since the update ships no Jinja template. Integrity 15, files in the deepseek-ai organisation with a SHA-256 for each shard. Safety evaluation 0, none published. Guidance for agent use 0, none found. Security programme 0, no disclosure policy found and security.txt answered 404 on deepseek.com and www.deepseek.com. 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. Reliability, 72 out of 100, up to 5.6 more on the total Why it scored 72: Read for open weights. Runtimes 12 of 30, vLLM's registry and mlx-lm's deepseek_v41 model (6 each). SGLang 0, its DeepSeek V4 file holds V4.1 parts but registers only DeepseekV4ForCausalLM (unchecked whether it loads V4.1). llama.cpp 0, no V4.1 architecture on master. Ollama 0, a cloud entry only. Reference implementation 20, the inference code DeepSeek publishes and updated on 12 October, though it isn't in Transformers 5.20.0. Sampling parameters published (15). Hardware 5 of 15, the card states no memory needs. The vendor's files are themselves a quantised build, FP8 and I8 tensors with BF16 for the rest (10). A general release, an update to the September model (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. ## 3. Schema & documentation, 69 out of 100, up to 5 more on the total Why it scored 69: Read for open weights. Card 9 of 25, it covers architecture, parameters and context but not intended use, limitations, languages or a knowledge cut-off (4 off for each). Chat template 10 of 20, the 17 October repository gives the prompt format only as code to run, encoding.py or the deepseek-recipe library. Tool calling 20, the DSML format documented in the encoding README with test cases, and DeepSeek's own deepseek-recipe parses it for any serving backend. Examples 10 of 15, encoding examples with tool calls and a torchrun demo, but no serving command for an API-style server on the card. Reasoning effort 1 to 100 and thinking mode documented (10). Versions 10, a dated, named update with a note of what changed and commit messages that say what each fixed. 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. 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). MIT, 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. ## 5. Agent ergonomics, 70 out of 100, up to 4.9 more on the total Why it scored 70: Read for open weights. Context 1M tokens, trained in (25). Tool calling it was built for, and the DSML format carries several invocations in one block (25). Structured output 0, not documented for the model. Reasoning control, an effort budget from 1 to 100 and a non-thinking mode (15). Fit 5 of 20, 510.3 GB of files, within 512 GB and well above 128 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. ## 6. Transparency & trust, 69 out of 100, up to 2.7 more on the total Made of editorial 65, provenance 73. Why it scored 69: Editorial half, read for open weights. Licence 30, MIT with nothing added. Training data 5 of 30, a general description (45T tokens of a multimodal corpus) with no sources and no cut-off. Evaluations 10 of 20, DeepSeek's numbers with the harness and sampling stated for each benchmark, and reproduction steps for DeepSWE v1.1 only. Version policy 20, each release in its own repository (V4-Flash, V4-Flash-0731, V4.1-Flash, V4.1-Flash-1017), and the update's card dates it and names what it replaces. 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): - Status page: 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 open weights. Last change 17 October 2026 (30). At least three dated changes in 90 days that say what they fixed, the chat template on 1 October, an indexer cache fix on 12 October, an encoding fix on 16 October and the 17 October update (20). Responsiveness 0, none of the 34 discussions listed for the September repository that aren't pull requests had a reply from a deepseek-ai member. Runtime support it maintains 15, the inference code and the deepseek-recipe library. File health 5 of 10, the update's card says it ships no Jinja template while the September repository now carries one, and an open pull request says that template is out of step with the 16 October encoding changes. 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: whether SGLang loads V4.1 by its architecture name; its DeepSeek V4 file holds V4.1 parts but registers only DeepseekV4ForCausalLM - unchecked: whether vLLM's DeepSeek V4.1 support is in a released version or only on main - The model doesn't fit our 128 GB machine at any precision DeepSeek or a runtime publishes, so our local runs may not score it - Performance and Task success are pending until our local runs ## Weaknesses - 510.3 GB of files, with no memory needs stated by DeepSeek - No Jinja template in the 17 October repository, so prompts are built by running DeepSeek's encoding code - Not in Transformers 5.20.0 or llama.cpp, and Ollama serves it from its cloud only - No safety evaluation, training sources or knowledge cut-off published - None of the 34 discussions listed for the September repository had a reply from DeepSeek ## 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. - Build prompts with encoding.py or deepseek-recipe; the September repository's Jinja template is behind the 16 October encoding changes per an open pull request - Parse tool calls from the DSML `calls` block, which V4.1 writes with a space before `calls`, `invoke` and `parameter`, unlike V4 - Pass reasoning effort as a number from 1 to 100, or low, high or max (50, 75, 100), in thinking mode - Allow 256K or more output tokens at high effort, as the card recommends - Use DeepSeek's hosted API (the deepseek-api listing) if you can't hold 510 GB of weights ## What the review panel asked for - sync the template - answer discussions - stated memory needs - a smaller official build - ship a template with each release - a serving example - a Jinja template in every release - 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.