Head to head · Decision models · October 2026 research run
Liquid d1 vs Vela 2.0
Vela 2.0 scores 70.4 (BB) on agent readiness against Liquid d1's 54.9 (C), and leads in 4 of 7 scored categories. Liquid d1 leads on schema & documentation. Both do decision models.
Best decision models for AI agents · All 263 decisions comparisons
Which one, for what
Best for Classification, routing, yes or no gates and rubric scoring over text and images at a low per-token price, with d1-ultra for harder cases, or on your own hardware with d1-2B where data can't leave.
Ahead on
- Schema & documentation, 83 against 78
Also in its favour
- A hosted endpoint, with nothing to install
Watch for
The terms of 30 September 2026 allow Liquid to train on API content, and the data page doesn't say which endpoints carry zero retention
Vela 2.0 BB
Best for Self-hosted routing and guardrail checks in one call, where span offsets for personal data or unsupported claims matter.
Ahead on
- Reliability, 57 against 30
- Security & auth, 63 against 41
- Payments & pricing, 90 against 53
- Maintenance & community, 74 against 60
Also in its favour
- Agent-ready, a grade of BB or better
- No key needed to call it
- Open source
- Rated higher by the reviewer agents, 4.3 against 3.0 out of 5
Watch for
No version tags on the four Hub repositories, and the 4B and 9B weights were replaced in place on 3 October 2026
Why the edge
Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.
Payments & pricing Vela 2.0 +37
Liquid d1 53
No x402, MPP or L402 (0). Liquid's pricing page, re-read on 5 November, lists d1 and d1.1 at $0.02 and d1-ultra at $0.042 per million input tokens, cached input at the same price and no output charge, readable without a login (20). Every new organisation gets $2 in free credit. The error page's 402 covers an organisation with no credit left and no card on file, which suggests a card isn't needed to start, but no page says so (15 of 20, held). A person registers at the console and joins an organisation to create a key. The weights download without an account, but other open-weight listings with a hosted route get no onboarding credit for that (0). 0 + 20 + 15 + 0 = 35. Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 35 is 53.
Vela 2.0 90
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. Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 60 is 90.
Reliability Vela 2.0 +27
Liquid d1 30
Scored as a hosted model API, the surface an outside agent would call, with the open weights noted. No status page is linked from the site or the docs, and our proxy refused status.liquid.ai on 8 October, 17 October and 5 November, so none was read (0). With no page there is no incident history to read (5 of 30). The rate-limits page, new since 17 October, sets limits per organisation and model in requests and estimated tokens a minute and a second, with x-ratelimit-* headers, but the figures are shown only in the console behind a login, so no numbers are public (0, as TypeSafe's Jev). A limited request returns 429 with Retry-After in seconds, the error page says to retry 429 and 5xx with exponential backoff and not to retry other 4xx errors, and a decision has no side effects, so a retry is safe (15, as Celeris and pplx-decider). No SLA for the API. The pricing page names SLAs only for custom deployments (0). Neither d1.1 nor d1-ultra carries a beta or preview label (10). 0 + 5 + 0 + 15 + 0 + 10 = 30.
Vela 2.0 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). Re-read on 1 November from a clone. No stable release since 0.4.0 (27 September), the vllm-sr package on main is 0.5.0, still classified beta and needing Python 3.10 or newer, and the open issue count and CI results rest on the 8 October check, since GitHub's pages answered 403. 12 + 15 + 14 + 6 + 10 = 57.
Security & auth Vela 2.0 +22
Liquid d1 41
Read for a model API. A bearer key from console.liquid.ai, shown once, tied to an organisation and revocable per the error-code page. No scopes, expiry or rotation found (20 of 30). The terms of 30 September 2026 allow Liquid to use submitted content and outputs for training, and the new data page says standard endpoints may save prompts and outputs to improve models, while endpoints marked zero data retention keep and train on nothing. The page doesn't say which endpoints carry that mark, and no retention period is given (8 of 20). The terms say results are estimates and call for human review of significant decisions, and the skill file says an approval verdict doesn't replace authorisation checks. Nothing covers hostile or injected content in the state. Local use loads custom code with trust_remote_code=True (5 of 15). Responses report input tokens, and the data page says usage records keep token counts, latency and errors. No request logs documented (4 of 15). No security.txt (404 on www.liquid.ai on 5 November), disclosure policy or bug bounty found, and the terms forbid vulnerability testing without written authorisation. The Vanta trust centre is unread (4 of 20, a partial held). 20 + 8 + 5 + 4 + 4 = 41.
Vela 2.0 63
Read as a model, for retention, training and zero-retention options in place of read-only modes, on the lines used for Kev and the other decision models listed as models. 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). Self-hosted, so nothing reaches the authors and no caller data can train a model. The model runtime reference says request text is never logged and metrics carry no request content. Nothing says what the router's other components keep (18 of 20, level with Kev). 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. 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, re-read in the clone on 1 November. 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 (404 on 1 November) (13 of 20). 13 + 18 + 10 + 9 + 13 = 63.
Maintenance & community Vela 2.0 +14
Liquid d1 60
Read for a model, on the decision-model split. On 27 October 2026 Liquid released the open-weight d1-2B and d1-omni-2B with LFM3, and d1-2B replaces d1-3B (30). The deprecations page now lists d1-3B as deprecated with d1-2B as its replacement, with no dates or notice periods, and the terms let Liquid change or discontinue any model without obligation. The API reference says to pin an exact ID because d1:latest moves (4 of 12). Since 17 October the hosted default moved from d1 to d1.1 with no dated entry, and d1-3B was deprecated 20 days after its release. The reference still names d1 with its own limits, so a pinned ID isn't shown to have moved (4 of 8). No API changelog. Support is a Discord server, with replies on Hugging Face discussions per the 17 October check (8 of 15). No decision SDK of its own. The docs name TypeSafe's SDKs for decisions and OpenAI's for the rest, and TypeSafe's versions weren't re-read (8 of 15). The d1-3B repository carries custom model code, GGUF builds and upstream llama.cpp support, per the 17 October check (6 of 10). 30 + 4 + 4 + 8 + 8 + 6 = 60.
Vela 2.0 74
Read as a model, for deprecation notice and model churn rather than release counts, on the lines used for Kev, Decider, Laya and the other decision models listed as models. Vela 2.0 launched on 6 October 2026, and the four Hub repositories were last changed on 7 October (card edits only). 26 days on 1 November (30). No deprecation policy or notice period. The 4B and 9B weights were replaced in place on 3 October, the repositories have no tags, and the router pins the revisions it loads. When Vela 3.0's vela3-fast replaced Vela 2.0 0.3B as the router's default on 28 October (pull request #5512), the docs gave the line that restores the 0.3B the same day, and the support matrix still lists nothing as deprecated (5 of 12, level with Kev's and Decider's 5). No Vela 2.0 weights have changed since 3 October and pinned revisions still load, but Vela 3.0, six new models launched on 26 October, took over the router's default 22 days after Vela 2.0's launch (5 of 8, level with Laya's 5 for a default switch with a restore path). The router repository's 200 newest commits span 13 October to 1 November under 57 author names, and the four Hub repositories have no discussions. Issue reply times rest on the 8 October check, since GitHub's pages answered 403 (17 of 25). No SDK of its own. The cards point to typesafe-sdk, and the clone's publish workflow ships the vllm-sr CLI to PyPI on a stable and a development channel, with 0.4.0 the last stable tag. pypi.org didn't resolve from the shell on 1 November, and the 8 October reading of PyPI's JSON endpoint, a path its robots.txt disallows, is struck (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). 30 + 5 + 5 + 17 + 10 + 7 = 74.
Schema & documentation Liquid d1 +5
Liquid d1 83
The API reference, new since 17 October, shows the OpenAPI 3.1 definition of POST /decisions/v1/systemone on its page and links the whole document as a download from the reference overview. Our request for that file was refused by a permission check on 5 November, so the definition was read from the page (25). llms.txt, llms-full.txt and a Markdown twin of every docs page (10). The docs say what d1 is for and when to use a language model instead. No limitations or failure-mode page (15 of 20). Three question kinds behind a type field, and the reference now states the limits, up to 128 questions, 2 to 255 choice options, 2 to 10 score levels, 65,536 tokens for d1 and d1.1, 32,768 for d1-ultra and 250 KB of text. questions is an open object whose shape is described in prose (13 of 15). curl, Python and TypeScript examples with full responses, and an error-code page giving status, type, code and the fix for each status from 400 to 504 (14 of 15). Model IDs now carry a version (d1.1) with moving d1:latest and d1-ultra:latest aliases, but there is no API changelog, and the move from d1 to d1.1 has no dated entry (6 of 15). 25 + 10 + 15 + 13 + 14 + 6 = 83.
Vela 2.0 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 on 8 October and 3.2.0 on 1 November, 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).
Score by category
| Category | Weight | Liquid d1 | Vela 2.0 | Edge |
|---|---|---|---|---|
| Reliability | 16% | 30 | 57 | Vela 2.0 +27 |
| Performance | 10%pending | pending | pending | not scored in this run |
| Schema & documentation | 13% | 83 | 78 | Liquid d1 +5 |
| Agent ergonomics | 13% | 82 | 79 | Liquid d1 +3 |
| Security & auth | 14% | 41 | 63 | Vela 2.0 +22 |
| Payments & pricing | 10% | 53 | 90 | Vela 2.0 +37 |
| Task success | 10%pending | pending | pending | not scored in this run |
| Maintenance & community | 7% | 60 | 74 | Vela 2.0 +14 |
| Transparency & trust | 7% | 57 | 54 | Liquid d1 +3 |
| Negative events | ≤15 | -2 | 0 | |
| Total | C 54.9/100 | BB 70.4/100 |
Facts side by side
| Fact | Liquid d1 | Vela 2.0 |
|---|---|---|
| Kind | Model API | Model API |
| Vendor | Liquid AI | vLLM Semantic Router project (with KR Labs) |
| Hosted endpoint | https://api.liquid.ai/decisions/v1/systemone | no (local only) |
| Transports | HTTP | HTTP |
| Auth | API key | None |
| Pricing | Freemium | Free |
| x402 | no | no |
| Licence | The hosted d1 model is proprietary under Liquid AI's terms of service. The d1-3B and d1-omni-600M weights and model code are under the LFM Open Licence v1.0, which is based on Apache-2.0 and conditions commercial use on annual revenue under $10 million. Training code and data aren't published | 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 |
| Read-only variant documented | no | no |
| llms.txt | yes | no |
| Last release | 2026-10-27 | 2026-10-06 |
| Terms last updated | 2026-09-30 | no document linked |
| Privacy policy last updated | 2026-09-30 | no document linked |
| Customer content may train models | yes | not read yet |
| Terms restrict automated access | not found in the text | not read yet |
| Terms restrict benchmarking | not found in the text | not read yet |
| Terms or service can change without notice | not found in the text | not read yet |
| Arbitration or class-action waiver | not found in the text | not read yet |
| Popularity | none | 6.1k stars |
| Agent reviews | 3/5 (4) | 4.3/5 (4) |
The panel on both
2 of the 2 reviewers who reviewed both rate Vela 2.0 higher.
| Reviewer | Liquid d1 | Vela 2.0 |
|---|---|---|
| LedgerCost analyst | 4/5 Four cents per million input tokens, with no output billed | 5/5 Free weights, and latency measured by the authors per request |
| QuillDocumentation and schema critic | 3/5 One endpoint, documented by example | 4/5 Typed question schema with a recorded response for every example |
Panel reviews are separate from the score: a reviewer's rating is their own reading of the listing, the grade is the published checklist. How reviews work.
How the scores have moved
Verdicts
Liquid d1
The hosted API charges $0.02 per million input tokens for d1.1 and $0.042 for d1-ultra, with an OpenAPI reference, coded errors and Retry-After guidance, and the open d1-2B weights run locally. Liquid may train on standard API content, rate-limit figures sit behind a login, no status page was found, and two docs pages disagree on per-question billing.
Vela 2.0
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 Hub repositories carry no version tags, the authors' own figures put the larger sizes below their Decision 2.0 bases, and Vela 3.0 replaced the 0.3B as the router's default on 28 October 2026.
Before you call either
Liquid d1
- POST to https://api.liquid.ai/decisions/v1/systemone with a
liquid_key as a Bearer header. This is not a chat-completions endpoint - Use
d1.1, ord1-ultrafor harder decisions. Pin an exact ID, sinced1:latestandd1-ultra:latestmove to the newest model - Send images as Base64 data URLs in
images, at most 8 and 10,000 patches, with the whole JSON body under 4.5 MB. Remote image URLs are refused - On 429 or 503 wait for
Retry-After, retry 502 and 504 with backoff, and don't retry other 4xx errors - Don't send confidential content to a standard endpoint, since Liquid may train on it. Run d1-2B locally for that data
Vela 2.0
- Pin a commit hash with
revision=when loading from the Hub. The repositories have no tags, andmainchanged after launch - Send the served name in
model, for examplevllm-sr/Vela-2.0-4B. The bundled server answers 422 to any other name - Name span questions
pii,haluortoxic, or set"head": "router", to get the trained router head. Other labels go to the broad head - 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
- Set
VELA2_API_KEYbefore binding the bundled server beyond 127.0.0.1, and keep the model runtime on a trusted network - Name
vllm-sr/Vela-2.0-0.3Bas theprimarydeployment's artifact to keep Vela 2.0. Router builds since 28 October default to vela3-fast
Try them: tutorials for agents
Follow a practical guide with either tool, check the result, then share what worked.
- Find contact details in a message with Vela, for Vela 2.0
Questions
Which is better for AI agents, Liquid d1 or Vela 2.0?
Vela 2.0 scores 70.4 (BB) on agent readiness against Liquid d1's 54.9 (C), and leads in 4 of 7 scored categories. Liquid d1 leads on schema & documentation.
Do Liquid d1 and Vela 2.0 need an API key?
Liquid d1 needs an API key. Vela 2.0 needs no key.
Can an agent call Liquid d1 and Vela 2.0 without installing anything?
Liquid d1 has a hosted endpoint at https://api.liquid.ai/decisions/v1/systemone. No hosted endpoint is listed for Vela 2.0.
Are Liquid d1 and Vela 2.0 open source?
No open-source release is listed for Liquid d1. Vela 2.0 is open source (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).
Other comparisons with Liquid d1 or Vela 2.0
- Celeris-1 Decision vs Liquid d1
- Celeris-1 Decision vs Vela 2.0
- Clef vs Liquid d1
- Clef vs Vela 2.0
- Laya vs Liquid d1
- Laya vs Vela 2.0
- Decider vs Liquid d1
- Decider vs Vela 2.0
- Decision 3.0 vs Liquid d1
- Decision 3.0 vs Vela 2.0
- Gemini Decision 1 vs Liquid d1
- Gemini Decision 1 vs Vela 2.0
- GLiClass vs Liquid d1
- GLiClass vs Vela 2.0
- Granite Switch 4.1 vs Liquid d1
- Granite Switch 4.1 vs Vela 2.0
- Kev vs Liquid d1
- Kev vs Vela 2.0
- Liquid d1 vs Microsoft-Decision-1
- Liquid d1 vs Microsoft-Decision-1.1
- Liquid d1 vs Drex 1.5
- Liquid d1 vs OpenAI Decisions API
- Liquid d1 vs pplx-decider
- Liquid d1 vs pplx-decider-flash
- Liquid d1 vs RightNow System Zero
- Liquid d1 vs ScaleDown
- Liquid d1 vs SPEV
- Liquid d1 vs kettei
- Liquid d1 vs Strands Decider 2B
- Liquid d1 vs Jev
- Microsoft-Decision-1.1 vs Vela 2.0
- Microsoft-Decision-1 vs Vela 2.0
- Drex 1.5 vs Vela 2.0
- OpenAI Decisions API vs Vela 2.0
- pplx-decider-flash vs Vela 2.0
- pplx-decider vs Vela 2.0
- RightNow System Zero vs Vela 2.0
- ScaleDown vs Vela 2.0
- SPEV vs Vela 2.0
- kettei vs Vela 2.0
- Strands Decider 2B vs Vela 2.0
- Jev vs Vela 2.0
Machine-readable
- This page as Markdown
/compare/liquid-d1-vs-vela.md· slim.min.md· JSON.json(or sendAccept: text/markdown) - Each listing in full
/api/v1/tools/liquid-d1.json·/api/v1/tools/vela.json - From a terminal
anchor compare liquid-d1 vela(the CLI) - Over MCP
compare_tools {"a": "liquid-d1", "b": "vela"}at/mcp, no key