d1-ultra
Released Oct 15, 2026 · knowledge cutoff unpublished
- Status
- Unknown
- Location
- United States
- Type
- System One
- Modality
- Multimodal
- Context window
- —
- Max output
- —
- Speed
- —
- Price ($/MTok in / out)
- $0.04 / $0.00
- Cost per task
- —
- Open weights
- No
Strengths & weaknesses
Strengths
- Trained specifically to make its probabilities trustworthy, where Liquid says most decision models are overconfident — the point is a threshold you can automate on
- Built for messy visual judgement calls such as spotting a defect against a spec or checking a damage claim against customer photos
- Each question is billed as its own prompt on input tokens only, so cost scales with how much you ask rather than with any generated output
- Reachable through Liquid's own API, OpenRouter and Vercel AI Gateway on the same System One request shape as Jev
Weaknesses
- Closed weights, unlike Liquid's open d1 checkpoints — it runs only on hosted endpoints
- The lead over rival decision models is Liquid's own claim, and one independent review could not confirm it beats every rival
- Within two weeks Liquid was steering cost-sensitive users to a cheaper hosted d1.1, so the hosted line-up is still shifting
Not an LLM: Liquid AI's flagship hosted "System One" decision model, built on a larger open-weight base model with Liquid-specific customisations, taking text and/or images plus typed questions (yes/no, choice, score) and returning probabilities with no generated tokens, so none of this site's benchmarks apply and every cell is null by design. Filed under the fast tier with the other decision models. Launched 2026-10-15 (liquid.ai/blog/d1-ultra) at $0.042 per M input tokens and $0 output per Liquid's pricing page, with images billed at 1.5 tokens per 32x32-pixel patch. Liquid's launch thread claims 44 languages, first place on the Artificial Analysis Decision Index (65) and a lead in all eight categories of its own 112-evaluation suite against d1 and Jev 1.14; AA's Decision Index is not a tracked benchmark here, and these are Liquid's figures. Liquid publishes no context length in the excerpts seen; OpenRouter and Vercel list 32,768 tokens, so contextWindow is left null pending an official figure. OpenRouter's liquid/d1-ultra is the only API id verified; Liquid's own model string was not seen. Weights are not published. No knowledge cutoff or Artificial Analysis cost/speed entry. Sources are search excerpts, not pages read directly.
Benchmarks0 of 10 reported
MMLU-Proretired
GPQA Diamond
SWE-bench Verified
SWE-bench Pro
Terminal-Bench 2.1
AIME (latest)retired
Humanity's Last Exam
LMArena Elo
GDPval-AA v2
ARC-AGI-2
For developers
API model strings
- openrouter
liquid/d1-ultra
Licence
Proprietary — weights not released
Retirement
No retirement announced
Lineage
No recorded predecessor