# Vela 2.0 > Vela 2.0 is a family of four open-weight decision models from the vLLM Semantic Router project and KR Labs, released on 6 October 2026 under Apache-2.0 for routing, safety checks, personal-data spans and hallucination checks. - Canonical: https://www.anchorterminal.com/tools/vela - Markdown: https://www.anchorterminal.com/tools/vela.md (~6,900 tokens) - Slim: https://www.anchorterminal.com/tools/vela.min.md (~1,730 tokens, same facts, less prose, for token-sensitive contexts) - JSON: https://www.anchorterminal.com/tools/vela.json (this page as data, same URL with Accept: application/json) - Site index for agents: https://www.anchorterminal.com/llms.txt (full text: https://www.anchorterminal.com/llms-full.txt) - API: https://www.anchorterminal.com/api/v1/index.json - Updated: 2026-10-08 ## Overview **Grade B · 66.5/100 · rank #219 of 629 · #4 in Decision models · not agent-ready · confidence medium** ## Assessment One self-hosted call answers routing, prompt-attack, personal-data and unsupported-claim questions with probabilities and character offsets, under Apache-2.0 with SHA-256 manifests. The models are days old and carry no Hub version tags, and the three larger sizes keep 74 to 89 per cent of their Decision 2.0 bases on the Jev Decision Index by the authors' figures. ## Facts | Field | Value | | --- | --- | | Vendor | vLLM Semantic Router project and KR Labs (https://vllm-sr.ai) | | Kind | Model API | | Category | Decision models (https://www.anchorterminal.com/categories/decision-models) | | Transport | HTTP | | Auth | None · No account. The weights are public and ungated on Hugging Face. The bundled `vela2_serve.py` binds to 127.0.0.1 and ignores the Authorization header unless `VELA2_API_KEY` is set, after which it requires `Authorization: Bearer ` and answers 401 otherwise. The router's model runtime has no authentication and publishes on 127.0.0.1 unless `--host` is passed. | | Pricing | Free (Free · OSS) · Free and open source, with nothing to buy and no hosted API. Hardware is the owner's cost. The 0.3B runs on a CPU, and the cards put GPU parameter memory at about 17 GB for the 4B. The router docs list about 32 GB for the 9B (checked 2026-10-08). | | x402 | No · No x402, MPP or L402. Vela 2.0 is software the owner runs, and neither server has a payment route (checked 2026-10-08). | | Licence | Apache-2.0 (weights, code and documentation). The 0.3B's tokeniser keeps the Gemma Terms of Use, and training data keeps its own licences | | Packages | pypi: `vllm-sr` | | Source | https://github.com/vllm-project/semantic-router | | Docs | https://huggingface.co/collections/vllm-sr/vela-20 | | llms.txt | not found | | Last release | 2026-10-06 | | GitHub stars | 6,054 (as of 2026-10-08) | | Models | Vela-2.0-0.3B (307M encoder, from Decision-1.0-Kai), Vela-2.0-0.8B (756M), Vela-2.0-4B (4.2B) and Vela-2.0-9B (7.9B), the last three fine-tuned from Decision 2.0 Eos, Nox and Lux on Qwen3.5 backbones | | Licence | Apache-2.0 for weights, code and documentation. The 0.3B's tokeniser keeps the Gemma Terms of Use. Training data isn't redistributed and keeps its own licences, some CC BY-SA | | Question types | choice (2 to 255 options), noul (yes or no), score (2 to 10 ordered levels), set (any number of labels) and span (labelled character offsets), any mix in one request | | Context | 8,192 tokens on the 0.3B and 16,384 a rendered sequence on the decoders. Span targets over 2,048 tokens are read in windows of up to 1,800 | | Hardware | The 0.3B runs on a CPU or through ONNX. GPU with bf16 autocast is the evaluated setting for the decoders, about 17 GB of parameter memory for the 4B. fp16 isn't supported | | Serving | `AutoModel.from_pretrained(..., trust_remote_code=True)` and `model.system_one(...)`, the bundled `vela2_serve.py` on `POST /v1/systemone`, or the router's model runtime on `POST /v1/decisions` (development channel) | | Span heads | A router head trained on PII (17 types), unsupported claims and toxic spans, and on the decoders a broad head for open labels. The response names the head that answered | | Errors | Bundled server 401, 413 and 422 with `detail`. Model runtime 400, 404, 413, 422, 429 and 503 with `{error: {code, message}}` | | Hosted option | None | | Languages | 17 listed on the cards, among them Arabic, Chinese, English, French, German, Hindi, Japanese, Korean and Spanish | | Capabilities | inference.decision, guard.pii, guard.injection, guard.moderation, guard.self-host | | Tags | model, open-source, open-weights, self-hosted, local, free, python, openapi | | JSON | https://www.anchorterminal.com/api/v1/tools/vela.json | ## Score breakdown (methodology v0.4, October 2026 research run) Assessed 2026-10-08 from public evidence against the published checklist (https://www.anchorterminal.com/benchmark/#checklist). Confidence: medium. Performance and Task success pending (no score, not in the total); the total is Σ(score × weight) ÷ 80 over the 7 assessed categories. "This run" is each category's share of the 100 points. | Category | Weight | This run | Score (0–100) | Points | | --- | --- | --- | --- | --- | | Reliability | 16% | 20 | 57 | 11.4 | | Performance | 10% | pending | pending | n/a | | Schema & documentation | 13% | 16.2 | 78 | 12.7 | | Agent ergonomics | 13% | 16.2 | 79 | 12.8 | | Security & auth | 14% | 17.5 | 60 | 10.5 | | Payments & pricing | 10% | 12.5 | 60 | 7.5 | | Task success | 10% | pending | pending | n/a | | Maintenance & community | 7% | 8.8 | 84 | 7.3 | | Transparency & trust (editorial 68, provenance 27) | 7% | 8.8 | 48 | 4.2 | | Negative events | up to −15 | up to −15 | none recorded | 0 | | **Total** | | | | **66.5 → B** | ### Why each score - Reliability 57: Scored on the local-package checklist, since Vela 2.0 is a model the owner runs. There are two routes. The model cards load it with `pip install torch "transformers>=5.17"` and `AutoModel.from_pretrained(..., trust_remote_code=True)`, with no Python version stated. The router's `vllm-sr` package on PyPI states Python 3.10 or newer, but serving a model with it is newer than the 0.4.0 stable release and needs the development channel (12 of 20). The semantic-router repository has public CI whose model-runtime suite includes five `test_vela2_*` files, and each model repository ships a parity record against the research scorer with zero decision differences on 230 rows. We couldn't read today's CI result on main, since the GitHub API was rate-limited and the Actions page we got was stale (15 of 25). The router repository showed 408 open issues on 8 October, 23 of them mentioning Vela, against 200 commits between 23 September and 8 October. The four Hub repositories have no discussions. Issue #4668 of 7 October tracks the 0.3B's CPU latency, which the release note puts at 79 ms against 16 ms for Vela 1.0 (14 of 25). The Hub repositories have no version tags, and the 4B and 9B weights were replaced in place on 3 October. The router pins the revisions it loads and publishes a dated release note (6 of 15). Released as 2.0 with no beta label on the cards, though the `vllm-sr` package is classified beta and the family is two days past launch (10 of 15). - Performance: Pending. Latency is measured per call by our probes, which haven't run yet, so this run doesn't score it. Its weight is shared across the assessed categories until the first probe window closes. - Schema & documentation 78: Read for a model the owner serves. The router's model runtime carries an OpenAPI 3.0.3 file, contract version 2.2.0, served at `GET /openapi.yaml`, which covers `/v1/decisions` and its `/v1/systemone` alias. The server bundled in the model repositories has no specification file (22 of 25). vllm-sr.ai has no llms.txt (404). The model cards, USAGE.md, EVALUATION.md and TRAINING.md are Markdown in each repository (5 of 10). Each card states the size's intended place and its evaluation limits, and the router's release note names the signals where the 0.3B is behind Vela 1.0. We found no section on uses to avoid (15 of 20). Five question types with 2 to 255 options, 2 to 10 score levels and 1 to 255 labels, and enums for span head and long-input policy. State is free-form text or named parts (13 of 15). Recorded responses for every example, and error tables for both servers (400, 404, 413, 422, 429 and 503 with codes on the runtime) (14 of 15). The runtime contract is versioned and reports `api_version`, and the router has dated release notes. The models themselves have no tags or changelog (9 of 15). - Agent ergonomics 79: Read as an API an agent calls for a decision, as with Jev, Clef and Kev. Answers are probabilities and spans, `output_tokens` is always 0, and the decoders read the state once and fork it per question. Input is 16,384 tokens a sequence, or 8,192 on the 0.3B (20 of 25). The caller fixes the output shape with any mix of the five types, a `threshold` per question, a `head` override for spans and `/v1/bundle` for several tasks in one call (18 of 20). The runtime answers with `{error: {code, message}}`, and a failed question carries its own code without failing the others. The bundled server answers 422 with `loc`, `msg` and `type`, and 413 for oversize questions (16 of 20). Calls are stateless and safe to retry. The runtime answers 429 `overloaded` past 256 queued requests a model. We found no `Retry-After` header in the runtime code and no retry guidance (15 of 20). `typesafe-sdk` works against the bundled server per USAGE.md, and loading is one `AutoModel` call. Setup needs `trust_remote_code`, about 17 GB of GPU memory for the 4B, or the router's development channel (10 of 15). - Security & auth 60: Read as software the owner runs. No account. The bundled `vela2_serve.py` binds to 127.0.0.1 and is open unless `VELA2_API_KEY` is set, after which it requires one bearer key, compared in constant time. The router's model runtime has no authentication, and its docs say to expose it only on a trusted network (13 of 30). A decision model has no write actions (15 of 20). The models are trained to flag prompt attacks, each question sees only the state and its own block, and the router's SECURITY.md names adversarial input to classification models in its threat model. Nothing documents how hostile text in the state can move an answer (10 of 15). The runtime exposes Prometheus metrics and, on request, the revision, digest and timings that answered a call, and its docs say request text is never logged. The bundled server keeps no record of calls (9 of 15). SECURITY.md takes private reports through GitHub Security Advisories, with severity classes and a disclosure policy. Weights are safetensors with a SHA256SUMS manifest in the decoder repositories, and the runtime checks every file's SHA-256 and never runs code from a model repository. The `transformers` route does run repository code through `trust_remote_code=True`, with no tag to pin. No bug bounty, and no security.txt on vllm-sr.ai (13 of 20). - Payments & pricing 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. - Task success: Pending. Task success needs the category task suites run through each tool, which haven't run yet, so this run doesn't score it. Its weight is shared across the assessed categories until then. A data provider's data-quality score is published on its listing now and becomes half of this category when it's scored. - Maintenance & community 84: Read for an open-weight model. Launched on 6 October 2026, with the four model repositories last changed on 7 October (30). Vela 1.0 on 18 September, the first Vela 2.0 upload on 29 September, re-exported weights on 3 October and the launch on 6 October (20). The router repository's 200 newest commits span 23 September to 8 October from more than eight authors, and the release note links issues the maintainers opened in the same week. 408 issues were open and we didn't check reply times (17 of 25). No SDK of its own. The cards point to `typesafe-sdk`, and the `vllm-sr` CLI on PyPI had a development build dated 8 October (10 of 15). The router repository runs CI, a security scan and a package check, with today's result on main unread (7 of 10). - Transparency & trust 48: Apache-2.0 for the weights, code and documentation, on Apache-2.0 Decision 2.0 and Qwen3.5 bases, with MODIFICATIONS.md and ATTRIBUTIONS.md in each repository. The 0.3B's tokeniser keeps the Gemma Terms of Use, and training data isn't redistributed and keeps its own licences, some CC BY-SA. We found no training code (26 of 30). Self-hosted, so inputs stay on the owner's hardware. The runtime docs say request text is never logged and metrics carry no request content. There's no privacy statement for the project (22 of 30). The router's support matrix defines a Deprecated class and says nothing is in it, and the release note gives the lines that restore each Vela 1.0 model. No policy with notice periods (10 of 20). We found no telemetry statement either way and didn't audit the runtime's code for it. Weights download from the Hugging Face Hub (10 of 20). Fix list for a coding agent, everything this grade says the listing lacks, the biggest gain first (19 items): https://www.anchorterminal.com/fixes/vela.md (JSON https://www.anchorterminal.com/fixes/vela.json) ### What we couldn't check - unchecked: today's CI result on the semantic-router main branch. api.github.com answered 403 for the rate limit and the Actions page we were served was stale - unchecked: reply times on issues and pull requests in vllm-project/semantic-router - unchecked: PyPI download counts for `vllm-sr`. pypistats.org answered 429 - unchecked: whether the model runtime or the `vllm-sr` CLI sends any telemetry. We found no statement and didn't audit the code - The accuracy, calibration and latency figures are the authors', on their own harness. We haven't run them - The listing covers Vela 2.0 only. Decision 2.0 and Vela 1.0 sit in the same Hugging Face organisation and aren't graded here - The 0.3B repository's LICENSING_STATUS.md is inherited from Decision-1.0-Kai and says no runtime code is bundled, while the repository does ship `vela2_inference.py` - No legal entity was found. The site footer names the vLLM Semantic Router Team, and KR Labs is credited as co-lead on the model cards ### Sources - launch post, 6 October 2026: (seen 2026-10-08) - Vela 2.0 collection: (seen 2026-10-08) - Vela-2.0-4B model card: (seen 2026-10-08) - Vela-2.0-4B usage guide, with the server's protocol table: (seen 2026-10-08) - Vela-2.0-4B evaluation and disclosures: (seen 2026-10-08) - Vela-2.0-4B training stages and data licences: (seen 2026-10-08) - bundled server source: (seen 2026-10-08) - repository metadata, files, commits, tags and discussions for the four models: (seen 2026-10-08) - 0.3B licence scope and tokeniser terms: (seen 2026-10-08) - semantic-router repository (cloned), SECURITY.md, workflows and model-runtime tests: (seen 2026-10-08) - release note on the 0.3B becoming the built-in signal default: (seen 2026-10-08) - model runtime reference, HTTP API, errors, metrics and security: (seen 2026-10-08) - model runtime OpenAPI file: (seen 2026-10-08) - model runtime quickstart and release channel note: (seen 2026-10-08) - vllm-sr package metadata and release list: (seen 2026-10-08) - open issues: (seen 2026-10-08) - security.txt (404): (seen 2026-10-08) - domain registration: (seen 2026-10-08) ## Who's behind it (provenance 27/100, checked 2026-10-08) | Check | Finding | Points | | --- | --- | --- | | Legal entity named | not found | 0/20 | | Domain age | vllm-sr.ai, registered 2026-07-13 (under a year) | 0/15 | | Endpoint on the vendor's domain | no hosted endpoint | n/a | | Terms of service | nothing hosted, so the Apache-2.0 (weights, code and documentation). The 0.3B's tokeniser keeps the Gemma Terms of Use, and training data keeps its own licences licence stands in | 10/10 | | Privacy policy | nothing hosted, not scored | n/a | | Status page | not found | 0/10 | | Changelog | published | 10/10 | | security.txt | not found | 0/10 | An open-source project with no company named as publisher. The site footer reads vLLM Semantic Router Team, and the model cards credit KR Labs and vLLM Semantic Router. RDAP gives 13 July 2026 as the registration date of vllm-sr.ai. Software the owner runs, so there's no hosted endpoint, terms or privacy policy. The Apache-2.0 licence stands in for terms. vllm-sr.ai/.well-known/security.txt returns 404. SECURITY.md in the repository takes private reports through GitHub Security Advisories. The weights are on huggingface.co under the vllm-sr organisation, and the code for the router and its model runtime is at github.com/vllm-project/semantic-router. ### Terms and privacy, as read A reading by a fixed set of rules, each answered with the vendor's own sentence. Not legal advice. **Terms of service**. Nothing is hosted by the vendor, so there are no terms of service to read. The Apache-2.0 (weights, code and documentation). The 0.3B's tokeniser keeps the Gemma Terms of Use, and training data keeps its own licences licence stands in and the check scores in full. **Privacy policy**. Nothing is hosted by the vendor, so there is no privacy policy to read and the check isn't scored. ## Live (updated 2026-10-08 18:25 UTC) - github `vllm-project/semantic-router` v0.4.0, released 2026-09-27 - pypi `vllm-sr` 0.4.0, released 2026-09-27 - security.txt: none - Watching changelog - Always current: https://www.anchorterminal.com/api/v1/live/vela.json ## Probe metrics Not measured yet. Our benchmark probes haven't run, so there's no availability, latency or error rate from a run and Performance is pending. Live uptime, where we poll the endpoint, is under Live and doesn't change the score. ## Strengths - Five question types in one request (choice, noul, score, set and span), with span answers as labelled character offsets and a probability each - Apache-2.0 weights, code and documentation, ungated on Hugging Face, with safetensors files and a SHA256SUMS manifest in the three decoder repositories - Two serving routes. A bundled FastAPI server on `POST /v1/systemone`, and the router's model runtime with an OpenAPI 3.0.3 contract and Prometheus metrics - The model cards disclose evaluation protocol, including that the 0.3B release selection considered test results and that SQuAD v2 isn't zero-shot - The router's release note lists where the 0.3B default is behind Vela 1.0, with numbers, and how to restore each Vela 1.0 model ## Weaknesses - No version tags on the four Hub repositories, and the 4B and 9B weights were replaced in place on 3 October 2026 - Loading with `transformers` needs `trust_remote_code=True`, which runs Python from the model repository - The router's `vllm-sr serve MODEL` engine mode is newer than the 0.4.0 stable release and needs the development channel - By the authors' figures the 4B scores 31.63 on the Jev Decision Index 0.2.1 against 42.55 for its Decision 2.0 base - The model runtime has no authentication, and the bundled server is open unless `VELA2_API_KEY` is set ## Before you call it (notes for agents) 1. Pin a commit hash with `revision=` when loading from the Hub. The repositories have no tags and `main` has changed since launch 2. Send the served name in `model`, for example `vllm-sr/Vela-2.0-4B`. The bundled server answers 422 to any other name 3. Name span questions `pii`, `halu` or `toxic`, or set `"head": "router"`, to get the trained router head. Other labels go to the broad head 4. Keep input under 16,384 tokens a sequence (8,192 on the 0.3B). The bundled server answers 413 when the questions alone don't fit 5. Set `VELA2_API_KEY` before binding the bundled server beyond 127.0.0.1, and keep the model runtime on a trusted network ## Connect Install: ```bash pip install torch "transformers>=5.17" safetensors tokenizers numpy fastapi uvicorn # from a local snapshot of vllm-sr/Vela-2.0-4B python vela2_serve.py --model . --device cuda --port 8001 ``` First request: ```bash curl -s localhost:8001/v1/systemone -H 'content-type: application/json' \ -d '{"model":"vllm-sr/Vela-2.0-4B","state":"My card was charged twice and the parcel never arrived.","questions":{"issues":{"type":"set","instructions":"Which issues does the customer report?","criteria":{"billing":"payments, charges, refunds or invoices","shipping":"delivery of an order or a parcel","login":"signing in, passwords or account access"}}}}' ``` ## Similar tools Ranked by shared capabilities, then score. Same-category tools with no shared capability key are listed last. | Tool | Grade | Score | Rank | Shared capabilities | x402 | Markdown | | --- | --- | --- | --- | --- | --- | --- | | NVIDIA NeMo Guardrails | B | 68.4 | 169 | guard.injection, guard.pii, guard.moderation, guard.self-host | no | https://www.anchorterminal.com/tools/nemo-guardrails.md | | Lakera Guard (Check Point AI Guardrails) | C | 59.6 | 374 | guard.injection, guard.pii, guard.moderation, guard.self-host | no | https://www.anchorterminal.com/tools/lakera-guard.md | | Guardrails AI | D | 49.6 | 527 | guard.injection, guard.pii, guard.moderation, guard.self-host | no | https://www.anchorterminal.com/tools/guardrails-ai.md | | Google Cloud Model Armor | BB | 77.9 | 15 | guard.injection, guard.pii, guard.moderation | no | https://www.anchorterminal.com/tools/google-model-armor.md | | Amazon Bedrock Guardrails | BB | 74.8 | 55 | guard.injection, guard.pii, guard.moderation | no | https://www.anchorterminal.com/tools/amazon-bedrock-guardrails.md | | Presidio | B | 66 | 228 | guard.pii, guard.self-host | no | https://www.anchorterminal.com/tools/microsoft-presidio.md | ## Panel reviews (0) Reviewed by the Anchor panel (https://www.anchorterminal.com/reviewers/index.md): . Desk reviews, written from public documentation, pricing, terms, source and status history on 1 October 2026. No calls made. For a desk review, the outcome says whether the reviewer's questions could be answered from public material: success, partial or failure. How reviews work: https://www.anchorterminal.com/reviews/how-it-works.md ## Notable - Four sizes share one request format. Vela-2.0-0.3B is a 307M encoder for CPU and ONNX with 8,192 tokens of input, and the 0.8B, 4B and 9B are Qwen3.5-based decoders with 16,384 (source: ) - Not a TypeSafe product. `system_one(state, questions)` accepts the TypeSafe SystemOne request, and USAGE.md shows `typesafe-sdk` pointed at the bundled server (source: ) - By the authors' figures the 9B scores 41.09 on the Jev Decision Index 0.2.1 against 46.23 for its base Decision-2.0-Lux-9B, and the 4B 31.63 against 42.55. These are the authors' numbers, not ours (source: ) - The authors report a macro AUC of 0.871, 0.875, 0.921 and 0.921 for the four sizes over 14 public safety and prompt-attack sets, on trained task families (source: ) - The router made Vela-2.0-0.3B its built-in signal default on 7 October 2026, and its release note says the 0.3B is behind Vela 1.0 on modality, user feedback, domain and fact check, and takes 79 ms against 16 ms on a 12-core CPU (source: ) - The GitHub stars (6,054) are for the semantic-router repository. The four model repositories showed 371, 87, 67 and 75 downloads on 8 October 2026 (source: ) - The 0.3B card records that its final release selection also considered test results, and that its shipped PII threshold floor was lowered after the test effect was seen (source: ) ## Compare - [Clef vs Vela 2.0](https://www.anchorterminal.com/compare/cloudflare-clef-vs-vela.md): B 66.1 vs B 66.5 - [Laya vs Vela 2.0](https://www.anchorterminal.com/compare/convai-laya-vs-vela.md): B 69.2 vs B 66.5 - [Kev vs Vela 2.0](https://www.anchorterminal.com/compare/jaredpalmer-kev-vs-vela.md): B 67.4 vs B 66.5 - [Liquid d1 vs Vela 2.0](https://www.anchorterminal.com/compare/liquid-d1-vs-vela.md): E 43.5 vs B 66.5 - [OpenAI Decisions API vs Vela 2.0](https://www.anchorterminal.com/compare/openai-decisions-api-vs-vela.md): BB 71.5 vs B 66.5 - [Strands Decider 2B vs Vela 2.0](https://www.anchorterminal.com/compare/strands-decider-vs-vela.md): C 61.3 vs B 66.5 - [Jev vs Vela 2.0](https://www.anchorterminal.com/compare/typesafe-jev-vs-vela.md): B 62.1 vs B 66.5 ## Verify this listing For the vendor. The badge or a plain link to this page verifies the listing, from a page on vllm-sr.ai or one of its subdomains, or the README of github.com/vllm-project/semantic-router. It shows the listing is the vendor's and that the vendor knows it's here, and it never changes a grade, rank or review. The vendor sends the page's address to `POST https://www.anchorterminal.com/api/v1/verify` as `{"slug": "vela", "url": "…"}`, or calls the `verify_listing` tool at https://www.anchorterminal.com/mcp. We fetch the page once, then again every week; two failed checks in a row and the verification lapses, and a later pass restores it. What we check: https://www.anchorterminal.com/builders/index.md#verify HTML badge: ```html Vela 2.0 on Anchor Terminal ``` Markdown badge, for a README: ```markdown [![Vela 2.0 on Anchor Terminal](https://www.anchorterminal.com/badges/vela.svg)](https://www.anchorterminal.com/tools/vela) ``` Plain link: ```html Vela 2.0 on Anchor Terminal ``` ## Share this listing For the vendor. Sharing assets for social media, two PNGs of 1200 × 630 that say Vela 2.0 is listed on Anchor Terminal, with the vendor's logo and this page's address and no grade or score. - Dark: https://www.anchorterminal.com/assets/share/vela-dark.png - Light: https://www.anchorterminal.com/assets/share/vela-light.png