What happened
Nace AI’s model reference lists Drex Max v1.1 with API ID drex-max-v1.1 and a release date of October 28, 2026. It lists a price of $0.03 per million input tokens and a 262,144-token input limit for the whole request. The catalog also lists the original Drex Max ID, drex-max, as retired on November 8, 2026, and says it is served by Drex Max v1.1.
The retirement does not mean requests using the old ID fail. Nace says a retired model ID has no fleet of its own: a request naming it is answered by a newer version at that version’s price, and the response reports the newer version. Stored usage continues to show the retired ID, its name and its historical price. For the retired IDs in the catalog, Nace specifies that they use Drex Max v1.1’s request limits and $0.03 per million input-token price. New integrations should use the successor ID; the successor still has to be available to the caller’s API key.
The listed 262,144-token limit covers the entire Drex Max request, not each question independently. Nace says the shared budget includes state, images, all questions and their options, formatting, and decision markers. Adding questions therefore draws from one request-level budget. When state is long, the model reads a subset selected for the questions. This makes request construction relevant to capacity: increasing the number of questions or attaching images can reduce the remaining room for state and options.
The catalog distinguishes this behavior from Drex 2.0, whose limits apply per question. For that model, the state has its own limit, and the row limit covers that state plus the longest question; other questions do not add to that row limit. Nace also says all requests must meet question-count and body-size limits. The distinction means an integrator cannot assume that a limit expressed as 262,144 tokens has identical practical meaning across the catalog.
Nace also documents drex-latest as an alias currently pointing to drex-max-v1.1. The company says this alias moves to a newer version only on a date announced in advance; requests using it can therefore change model and price on that date. A versioned ID stays pinned while active, and Nace recommends changing the ID in code to move to another version. GET /v1/models returns the models permitted for the API key in that environment; Nace says access can vary by environment and key, including during a staged release. The CLI commands nace models and nace models --json use the same key and base URL.
What the archive adds
New Runtime covered Nace’s Drex Max v1 open-weights release on October 31, 2026, including its 27B multimodal model. This is a new operational fact on a known model story: the catalog now documents successor routing and ID retirement, rather than a new weights release.
Trend signal
This update supports the view that model choice is partly a deployment and control decision, not just a capability comparison. Nace’s distinction between version-pinned IDs, a moving alias, and retired IDs makes routing behavior part of the integration contract. It is a modest fit with harness architecture outlives model choice: teams need explicit controls around which model their application actually invokes.
Why now
The timing follows two dated steps in Nace’s catalog: Drex Max v1.1 was released on October 28, 2026, while the original drex-max identifier was retired on November 8. The documentation frames the later date as a routing transition, not a service cutoff. That sequence matters because an integration can remain operational after the retirement while silently moving to the newer serving version.
Why it matters
A successful API response does not guarantee that a caller is still using the model it originally named. After retirement, drex-max resolves to v1.1, and the effective price and request limits are those of that successor. Teams should therefore treat aliases and retired IDs differently from pinned version IDs in production configuration.
The catalog gives integrators a practical way to reduce ambiguity: use a versioned ID when stable model identity matters, inspect the response’s model field, and check GET /v1/models with the real API key and environment before rollout. These steps do not establish that v1.1 is behaviorally equivalent to the original Drex Max; they make the routing change visible and help teams manage it as a migration rather than assuming the old name preserves the old model.
This is especially relevant where a model call is embedded in classification, extraction, or decision workflows whose prompts, test sets, and downstream expectations were tuned to an earlier release. A legacy identifier keeps application code running, but it does not preserve a fixed serving fleet. Recording the returned model identifier alongside outputs makes later regressions or cost changes easier to investigate.
What happens next
- Teams using
drex-maxwill need to update integrations if they want a pinned model identity. (3 months, 75%). Watch for: Nace continuing to mark legacy IDs as retired and route them to successor versions.
Limits and uncertainty
The catalog documents IDs, prices, token limits, and routing behavior, but does not report a behavioral comparison between the original Drex Max and v1.1. It also does not establish whether every API key or environment can use v1.1: Nace explicitly says access can differ and may be staged. The stated input price and request limit describe the successor’s terms, not a benchmark of cost or quality on any particular workload.
Nace does not quantify how much state is read when a long state exceeds what the model uses for particular questions, nor does the catalog provide migration performance data for requests sent through retired IDs. Teams should confirm their own key’s model access, inspect actual response identifiers, and run representative evaluations before relying on behavioral equivalence.
Skill to build now
Add a model-identity check to the integration boundary this week. For a production workflow, send the intended versioned ID, store the response’s model value with the result, and alert when it differs from the configured expectation. Use GET /v1/models or nace models --json in the deployment environment before enabling a model for a key. Use a pinned ID when reproducibility, evaluation, or a fixed cost assumption matters; use drex-latest only when accepting a scheduled future version change is intentional.
Reader takeaway
The retired drex-max ID still answers, but it no longer identifies a model with its own fleet. If predictable model behavior matters, pin a versioned ID, verify key-level access, and record the model returned in responses rather than relying on a legacy name or moving alias.