What changed for Jev builders, 2 to 7 November 2026

TypeSafe's docs now list images as an input for both Jev versions and a Box question type that only Jev 1.14 answers. Its aliases, model IDs, price, signup and credit state did not change. Base Labs' three decision models get a row among the hosted alternatives, and the versions page gains a dated list of projects that stayed on Jev 1.13. This report gives each change with its source and time, what was re-read and found unchanged, and what was not checked.

TypeSafe's docs: images on both versions, Box on 1.14 only

The models page is TypeSafe's main reference for model IDs, prices and limits. Between the read of 2 Nov 2026 (11:13 UTC) and the read of 7 Nov 2026 (15:29 UTC) it changed in three places. The page gives no date for these edits. A new docs changelog page dates the Jev 1.14 features to 14 Oct 2026, when Jev 1.14 was released; its aliases, model IDs and prices did not change.

• What it means for a builder: both versions document image input, and Box questions (a bounding box in an image) work only on Jev 1.14. The pool sent no image and no Box question, so neither is tested here. Documented
• Languages: the models page no longer says that English is the primary training language. TypeSafe publishes no multilingual accuracy figures; test on your own content. Documented
• Cookbook figures (TypeSafe's own, Reported): the re-ranking cookbook now reports top-1 accuracy rising from 5% to 30% and top-10 from 38% to 65% (earlier 5% to 18% and 38% to 62%); the parallel-questions cookbook says "12.2x cheaper and 11.8x faster" (earlier 10.0x faster) (R27-S15, R27-S16).
• Moved docs pages: the Jev 1.13 "model jaggedness" page and the intent-routing pattern now redirect to model-jaggedness.md and patterns/probability-routing.md (R27-S17).

Updated with this wording: limitations, what is Jev, Jev vs decision APIs (the Jev input cell), versions and aliases, re-ranking, content moderation and the other pages that described Jev's inputs or question types. Older dated sentences stay as dated history.

Base Labs: three decision models on Baseten

Base Labs is Baseten's research lab (labs.baseten.co). Its mammoth-1, elephant-1 and joey-1 are served on Baseten's Model APIs and now have one family row among the hosted APIs from other providers. A different model; your data goes to Baseten under Baseten's terms.

• Data terms (Baseten, quoted): "For synchronous Model API requests, Baseten doesn't store inputs or outputs under its zero data retention policy" (docs; queued requests are handled differently). Terms 6.3: "Baseten will not use Customer Content, including Customer Models, Model Outputs, or inference logs, to train, fine-tune, or otherwise develop machine learning or artificial intelligence models." Terms 6.4 allows technical and operational telemetry ("Usage Information") (R27-S210, R27-S212). Documented
• Availability: live on Baseten Model APIs with per-token pricing and no preview label; Baseten's signup page is self-serve (seen, not used) (R27-S200 to R27-S202, R27-S208).
• Other route: OpenRouter has per-model endpoints baseten/mammoth-1, baseten/elephant-1 and baseten/joey-1 (provider BaseTen, 16,384-token context, the same per-token prices); they are not in OpenRouter's public model list (460 models on 7 Nov). OpenRouter's own data terms for them were not read (R27-S213 to R27-S218).
• Rank claims (Reported): Base Labs' cards say mammoth-1 is "#1 of 208 models on the Jev Decision Index 0.4: 71.24"; elephant-1 "tied #8 of 208"; joey-1 "#37 of 230" on the public split. Not matched against the board on 7 Nov 2026 (its data returned HTTP 500 at 15:31 UTC). The benchmarks line of 24 Oct, which compared the "#1 of 208" claim with the board's own data, stands (R27-S204 to R27-S206, R27-S220).

Alternatives now lists 39 substitutes (13 model weights, 11 local runtimes, 4 request adapters, 2 classifier or LLM approaches and 9 hosted APIs from other providers; 13 + 11 + 4 + 2 + 9 = 39); the Base Labs weights appear as a line inside the hosted row and are not counted as a weights row. Details: alternatives.

Jev 1.14 vs 1.13: who stayed on 1.13

On 7 Nov 2026 (15:34 UTC) Google completed "jev 1.14 vs" to "jev 1.14 vs 1.13"; Bing and DuckDuckGo gave no completion (R27-S221 to R27-S223). A suggestion shows that people search for a phrase, not how many do. The versions and aliases page now uses the phrase in its title and lists eight GitHub issues and pull requests, read on 7 Nov, that discuss staying on, pinning or leaving 1.13 (Reported): five pin or keep a default on 1.13, one moves its default to 1.14, and two are about alias resolution. Reasons given include a project benchmark that "fell from 91% to 85%" after jev-latest moved, thresholds tuned on 1.13, and hard-coded IDs (R27-S224 to R27-S233). Eight items are not a survey.

OpenCode's free ID: unchanged

Third-party route; not run by TypeSafe. On 7 Nov 2026 (15:33 UTC) OpenCode's docs still list jev-1.13-free at "Free" input and output, "available on OpenCode for a limited time", with no end date or quota, and jev-1.13 at $0.042 input with output free. The zero-retention sentence, its exceptions list (no Jev entry) and the Terms' unpaid-account clause (effective 15 Aug 2026) are unchanged. The docs address now resolves to opencode.ai/docs/console/models/ (R27-S37, R27-S38). OpenRouter still lists no Jev ID ending :free (R27-S19). Details: third-party routes.

What else changed

Re-read on 7 November and unchanged

• Access: signups open, no free credit for new accounts, API operational ("All services are online"; the status page's history lists no event dated 31 Oct to 7 Nov). The status rests on the CEO post, TypeSafe's docs and pages and the status page (R27-S01 to R27-S14). Details: access status.
• Models and price: both aliases give jev-1.14.0; jev-1.13.0 is listed with no deprecation date (also none on the changelog); $0.042 per million input tokens, output free; context limits and the per-organization rate-limit wording unchanged; TypeSafe's legal pages show the same "Last updated" dates (R27-S05, R27-S11, R27-S15, R27-S18).
• The six routes: TypeSafe, OpenRouter, Vercel, Opper, Netlify and Cloudflare list the same Jev IDs; Opper's typesafe/jev-latest still gives jev-1.13.0 (15:34 UTC; R27-S19 to R27-S23). Details: versions and aliases.
• Lookalike sites: the three sites named by Eye Security still answer, and the report still names them (R27-S14).
• Comparison cells: d1-ultra's price, Mouse-1, Drex, Eevo's terms cell, Microsoft-Decision-1's post, Learn page and OpenRouter cells, OpenAI's input and route wording, and SubQ Snap's status and price were read again and are unchanged (R27-S27 to R27-S34, R27-S47).

Not checked or not read

• Three older comparison cells could not be settled on 7 Nov and keep their dates: OpenAI's Decisions API retention and ZDR wording (12 Oct), Cloudflare's Clef data-usage cell (12 Oct) and SubQ Snap's retention wording (14 Oct).
• The contents of ohmyjev 0.14.0, langchain-typesafe 0.0.2, pydantic-ai-slim 2.66.0 and Laya v0.4.28; Kev; Eevo's prices; the Amazon Bedrock, Azure AI Foundry and Workers AI catalogs.
• OpenRouter's own data terms for the Base Labs endpoints, and the leaderboard data behind Base Labs' rank claims (HTTP 500).
• Image input and Box questions: documented by TypeSafe, not tested by the pool.
• Reddit, X, YouTube and TikTok could not be searched for the days between the checks. No request was sent to any route or provider, and no account was created.

Sources

Earlier dated reports: 31 Oct–2 Nov 2026, 31 Oct 2026; all reports are listed on Research.