Tasks and inputs
Drex 2.0 is listed for Structured Decisions. It accepts text and produces probabilities. Use the publisher’s input format and preprocessing for your chosen deployment.
Drex 2.0 scores typed questions over text and returns probabilities. Its hosted route uses 2.0 for short states and falls back to 1.5 for longer requests.
Pricing conditions: Drex 2.0 input costs $0.06/M tokens; output tokens are not billed. The hosted 2.0 fleet handles states up to 8,192 tokens and state-plus-question rows up to 16,384. Longer accepted requests fall back to Drex 1.5 at its $0.05/M input rate. The response identifies the serving version. Downloadable weights have separate configurable context limits.
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
Drex 2.0 is a decision model, released on 6 October 2026, that scores named questions against a supplied text state. It returns probabilities rather than generating a chat answer. Its downloadable backbone is a 35B mixture of experts with about 3B active parameters, under the Nace.AI Open RAIL-M licence. The vision tower is not used by this decision model.
The hosted API accepts a state budget up to 131,072 tokens, but the 2.0 fleet itself handles states up to 8,192 and state-plus-question rows up to 16,384. Longer requests are answered by Drex 1.5. The table’s larger accepted limits therefore must not be described as a 2.0-only serving allowance. Local weights have their own configurable request context.
drex-latest moved from 1.5 to 2.0 on 12 October 2026. Inspect the response model field to identify the version that actually scored the request. The 2.0 input rate is $0.06 per million tokens; a long request served by 1.5 uses its $0.05 rate. No output-token charge or chat completion output allowance is inferred from the probability interface.
Use held-out examples with known decisions and measure calibration as well as the chosen label. Include ambiguous cases and long states that exercise fallback. Keep question definitions and action thresholds versioned, and inspect false positives and false negatives at the threshold your application will use.
Record the exact model identifier, deployment route and request settings with each result. Count complete runs, including rejected outputs, retries and tool calls. When the workflow creates files or code, open the files or execute the existing checks before accepting the result. A plausible first response can hide an incomplete final deliverable.
The specifications describe supported interfaces and request capacity. They do not establish a measured quality ranking or a latency guarantee for your account. SyncDev has not independently benchmarked this model. Choose a deployment after testing the languages, input formats and operating conditions that matter to your users.
Explore nace-ai models, browse model tasks, or compare two models with the same acceptance criteria.
Compare the model on the work you need to complete.
Drex 2.0 is listed for Structured Decisions. It accepts text and produces probabilities. Use the publisher’s input format and preprocessing for your chosen deployment.
Documented limits: 8.19K context. The listed route has starting rates of $0.06 input and $0.00 output per million tokens. Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.
The weights are downloadable under Nace.AI Open RAIL-M. Check runtime support and size the complete model and working context for your expected concurrency.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
phocinae
Phocinae Largha 150M v1 from phocinae: Encoder decision model that answers typed questions (noul, choice, score) over text with calibrated probabilities, on CPU and GPU.
Liquid AI
Liquid d1 returns typed binary decisions, class distributions or scores through Liquid AI’s decision API.
Nace AI
Drex 1.5 evaluates typed questions against a shared state and returns calibrated probabilities through the System One API.
TypeSafe
TypeSafe’s Jev evaluates many questions against shared text state through a System One interface; the latest alias currently resolves to Jev 1.13.