Tasks and inputs
SubQ Snap 1.1 is listed for Text classification, Structured Decisions. It accepts text, image and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
SyncDev model directory · Profile updated Nov 18, 2026
SubQ Snap 1.1 is Subquadratic's October 2026 decision model update: sharply better calibration, 20% cheaper at $0.04 per million input tokens, free output.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Indicative: $0.04 per 1M input tokens, output free. Earlier versions are auto-upgraded to Snap 1.2 at the same price.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Calibration was the weak spot, and version 1.1 went straight at it. When Subquadratic shipped SubQ Snap 1.1 on 22 October 2026, ten days after the first release, the headline was not a higher accuracy score. It was that the model's probabilities started meaning something.
On held-out problem families, expected calibration error fell from 0.108 to about 0.014. Overconfident hallucinations on out-of-distribution questions dropped from 99.3% to 12.5%. For a model whose entire output is a probability, those are the numbers that decide whether you can set an auto-approve threshold.
The When2Call results show the same shift. Overall accuracy rose from 75.2 to 96.2, and correct declines, where the model recognises it should not pick an answer, went from 49.1% to 98.3%. In a routing pipeline, a confident refusal is worth more than a confident wrong label.
All self-reported, against TypeSafe's Jev 1.14 unless noted:
| Benchmark | Snap 1.1 | Comparison |
|---|---|---|
| Artificial Analysis Decision Index | 65 (tied first) | Jev 57 |
| HF Decision Index 0.2.1 | 65.7 | Jev 60.6; Snap 1.0 61.8 |
| BFCL | 96.6 | Jev 95.3 |
| Humanity's Last Exam | 19.7% | Jev 18.5% |
Across its 16-benchmark enterprise suite, 1.1 averaged 5.5 points above 1.0 and 7.2 points above Jev. Subquadratic also published one clear loss: on PhishNChips it trailed Jev by 22 points. If phishing or fraud triage is your use case, test that category specifically.
Input fell 20%, to $0.04 per million tokens, and output stayed free. The release notes describe a native 384K-token context of interleaved text and images, though the following 1.2 announcement says Snap models cover 256K; probe the real limit with your own long inputs. Access is through the API at console.subq.ai. Treat prices as indicative.
SubQ Snap 1.2, released 11 November, kept the same price and fixed a problem 1.1 still had: reorder the answer options and the top choice changed 16.0% of the time. In 1.2 that fell to 0.0%. Subquadratic moves earlier-version users to 1.2 automatically. So treat 1.1 as the release that made the probabilities trustworthy, and 1.2 as the one that made them stable.
A quick validation for either: shuffle the options on 100 of your own questions and count how many top answers flip. Then plot predicted probability against observed accuracy in ten buckets. Compare with Jev and others under structured decisions.
Compare the model on the work you need to complete.
SubQ Snap 1.1 is listed for Text classification, Structured Decisions. It accepts text, image and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Documented limits: 384K context. The listed route has starting rates of $0.04 input per million tokens. Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.
Use the model identifier and account requirements for the selected API route. Confirm its request limits and output format with one complete task before connecting a larger workflow.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
Microsoft
Microsoft-Decision-1.1 is a decision model for routing, triage and classification, returning calibrated probabilities. GA in Foundry from November 2026.
Liquid AI
Liquid AI d1.2 is a hosted omnimodal decision model for text, image and audio questions with calibrated answers, at $0.02 per million input tokens.
Liquid AI
Liquid AI d1.2-ultra is the top hosted decision model in Liquid's line, scoring 75 on the AA Decision Index, with audio support at $0.042 per 1M input.
Gemini Decision 1 powers Google's Decisions API, returning calibrated yes/no, choice and rubric answers instead of text, at $0.10 per 1M input tokens.
subquadratic
SubQ Snap 1.2 is Subquadratic's long-context decision model on linear sparse attention, tested to 256K tokens, priced at $0.04 per 1M input tokens.
Perplexity
pplx-decider-flash-v1.3 is Perplexity's small, half-price decision model, distilled from the 27B decider, for high-volume classification and span extraction.