Published
Updated
Liquid's chart puts d1 at 58.9 on an internal Decision Index 0.2.1
Liquid AI posted a chart on September 29 captioned as an internal reproduction of Decision Index 0.2.1. The index line is d1 58.9, Jev 1.13 57.9, difference +1.0. Arts, language, and retrieval are ahead on that chart. Tools and knowledge are behind. The docs name a hosted endpoint and the model id d1:free. We did not call it. On October 7, 2026 the JevBench API board ranks d1 second at composite 73.0, with capability 74.4. Opened October 8, 2026, d1 is still second at 73.0, and Jev 1.13.0 on that board is fourth at 71.5. The same day's open weights, d1-3B at 48.57, are a separate page.
Liquid AI posted on September 29 a chart captioned “Internal reproduction of Decision Index 0.2.1 by Hugging Face.” The index row, five areas, is d1 58.9, Jev 1.13 57.9, difference +1.0. The public kit treats scores within 0.25 as a tie, so +1.0 sits outside that band. The board file opened September 30 lists Jev at 57.91. This chart rounds that cell to 57.9. We did not re-download the file, and we did not search it for the string d1.
The area rows, d1 / Jev 1.13 / difference, are Arts 45.5 / 37.7 / +7.8 on 7 benchmarks, Language 67.6 / 62.0 / +5.6 on 10, Retrieval 60.7 / 55.4 / +5.3 on 6, Tools 74.1 / 75.1 / -1.0 on 5, and Knowledge 43.3 / 51.3 / -8.0 on 10. The counts add to 38, which matches edition 0.2.1. The knowledge gap on this chart is a different pair from Matilda’s 43.52 against 51.40.
A reply in the same minute links the docs, a migration guide, and Liquid4All/cookbook examples/road-decider. We did not open that tree. The docs page we opened names POST https://api.liquid.ai/decisions/v1/systemone, model d1:free, and keys prefixed liquid_. It says a TypeSafe SDK (typesafe-sdk or @typesafe-ai/sdk) can set base_url to https://api.liquid.ai. Output tokens are always 0. The response types named are Noul, Choice, and Score. A noul of 0.999 on that page is a documentation example. The sections we read do not print a context window, a parameter count, a paid price, or a license.
@heath0xFF asks in the thread whether the weights will be available to download and run locally. A search of posts from @liquidai for local, download, or weights returned no results. We did not open a reply that prints a date for weights. Calls go to Liquid’s API. Access on this desk does not list that host as a door for calling Jev.
On October 7, 2026 the JevBench API board, release v1.6.1 under revision v1.7.12, ranks d1 second at composite 73.0. The capability list prints 74.4, Intelligence 62.1, Calibration 86.8, $0.017 per 1,000 decisions, p50 0.27 s, capability rank 3. The v1.7.8 note says d1 answered all 1,500 items on October 6, is scored without equating, and names a tariff of $0.04 per million input tokens. Jev 1.13.0 on that board is composite rank 3 at 71.5. 58.9 stays the September 29 Decision Index chart. We did not rerun the board.
Opened October 8, 2026, the same API board is still release v1.6.1, revision v1.7.21, and ranks 26 hosted systems. d1 is still second at 73.0, and the capability card is still rank 3 at 74.4. Jev 1.13.0 is fourth at 71.5. We did not rerun the suite.
On October 7, 2026 Liquid posted open weights under the name Open d1. The d1-3B card prints 48.57 on a self-run of Decision Index 0.2.1. That 48.57 is a different figure from this chart’s 58.9 and from the JevBench composite of 73.0. The write-up is Open d1.
This site's reading
Editorial notes evaluating claims against primary sources, contextualizing findings alongside related implementations, and defining technical terms.
Verify
Primary post is @liquidai, September 29, 2026, 18:35 UTC. The attached chart is captioned "Internal reproduction of Decision Index 0.2.1 by Hugging Face." The index row, five areas, prints d1 58.9, Jev 1.13 57.9, difference +1.0.
Area rows, d1 / Jev 1.13 / difference. Arts, 7 benchmarks: 45.5 / 37.7 / +7.8. Language, 10: 67.6 / 62.0 / +5.6. Retrieval, 6: 60.7 / 55.4 / +5.3. Tools, 5: 74.1 / 75.1 / -1.0. Knowledge, 10: 43.3 / 51.3 / -8.0. Those counts add to 38, which is edition 0.2.1's benchmark count.
The public kit's tie band is 0.25. +1.0 is outside that band. The public board file opened September 30 lists Jev at 57.91. This chart rounds that cell to 57.9. We did not re-download the file on October 1, and we did not search it for the string d1.
The reply in the same minute links the docs, a migration guide, and https://github.com/Liquid4All/cookbook/tree/main/examples/road-decider. We did not open that tree.
The docs page we opened names POST https://api.liquid.ai/decisions/v1/systemone, model id d1:free, and keys prefixed liquid_. It says a TypeSafe SDK, typesafe-sdk or @typesafe-ai/sdk, can set base_url to https://api.liquid.ai. It says output tokens are always 0. The response types named are Noul, Choice, and Score. A noul of 0.999 on that page is a documentation example.
The sections we read do not print a context window, a parameter count, a paid price, or a license. The model id printed there is d1:free.
@heath0xFF asks in the thread whether the weights will be available to download and run locally. A search of posts from @liquidai for local, download, or weights returned no results. We did not open a reply that prints a date for weights.
On October 7, 2026 https://benchmarkheaven.com/jev-models/api names release v1.6.1 and revision v1.7.12. d1 is composite rank 2 at 73.0 and capability 74.4, Intelligence 62.1, Calibration 86.8, $0.017 per 1,000 decisions, p50 0.27 s. The v1.7.8 note says d1 answered all 1,500 items on October 6, is scored without equating, and names a tariff of $0.04 per million input tokens. We did not rerun that board.
On October 7, 2026 @liquidai posted Open d1, status 2107878924831379676. The d1-3B card, opened October 8, prints 48.57 on a self-run of Decision Index 0.2.1 and says the run was not submitted. That 48.57 is not this chart's 58.9, and it is not the JevBench composite of 73.0. The write-up is the Open d1 page.
Compare
The September 28 board file lists Jev 1.13.0 at 57.91 on Decision Index 0.2.1. Liquid's chart is a separate reproduction at 58.9 against 57.9. Nace's page prints Drex 1.5 at 58.28 on the same edition. Matilda's self-run prints 59.26. Those three numbers are three authors' runs.
Matilda's knowledge cell is 43.52 against Jev at 51.40. Liquid's knowledge cell is 43.3 against 51.3. The pairs are close and they are not the same table.
Calls on this endpoint go to api.liquid.ai. Access on this desk does not list that host as a door for calling Jev. JevBench's 72.1 is a different board. This chart does not print a JevBench score. The API board opened October 7 ranks d1 second at composite 73.0 and capability third at 74.4.
Open d1, posted the same day, is a pair of open checkpoints. The d1-3B card prints 48.57 on a self-run of Decision Index 0.2.1. That figure stays on the Open d1 page. This page stays the hosted chart and the hosted JevBench row.
Terms
- d1:free
- The model id on Liquid's decision-model docs. The page we read does not print a paid model id or a paid rate.
- 58.9
- d1's index on Liquid's September 29 chart, an internal reproduction of Decision Index 0.2.1. Jev 1.13 is 57.9 on the same row. The difference printed is +1.0.