Where to use Jev (TypeSafe AI) and what it costs: routes compared

You can call Jev directly from TypeSafe or through five gateways. The listed input price is the same or nearly the same on every route that shows one; the routes differ in request shape, context limit, data terms, billing and who you have an account with.

Choose a route

Route matrix

Documented Every field is a public listing or documentation page as displayed on 5 Oct 2026, 05:13 UTC, unless another date is shown; data terms were read on 1 Oct, 05:26–05:31 UTC. On 6 Oct 2026, 05:15–05:24 UTC the pool rechecked TypeSafe's models page (model, aliases, price, limits, context), OpenRouter (Jev 1.13 price, context and ZDR listing; the Jev Router listing), Vercel (models API, providers, model page), Cloudflare (price and ZDR label) and Opper (price, context, ZDR wording, aliases). The only change since 5 Oct was Vercel's provider list (Vercel note); on 3 Oct TypeSafe's request limit had changed from 40 to 80 per second. On 7 Oct 2026, 05:19 UTC the pool rechecked TypeSafe's models page, OpenRouter (Jev 1.13 price, context and ZDR listing; the Jev Router listing), Vercel (models API and providers), Cloudflare (price and ZDR label) and Opper (price, context, ZDR wording, aliases): no change (7 Oct research report). On 8 Oct 2026, 05:14 UTC the pool rechecked TypeSafe's models page (model, aliases, price, limits), OpenRouter (Jev 1.13 endpoint, context and ZDR listing; the Jev Router listing), Vercel (models API and providers), Cloudflare (price and ZDR label), Netlify (model IDs) and Opper (price, context, aliases): no change (8 Oct research report). No update was published on 9 Oct; the 9 Oct evidence notes recorded access unchanged. On 10 Oct 2026, 05:12 UTC the pool rechecked TypeSafe's models page (model, aliases, price, limits), OpenRouter (Jev 1.13 endpoint and ZDR listing; the Jev Router listing), Vercel (models API and providers), Cloudflare (price and ZDR label), Netlify (model IDs) and Opper (price, context, aliases): only OpenRouter's stated context changed, from 32,000 to 64,000 tokens with a 32,000-token prompt limit (10 Oct research report). Which model IDs and aliases each route accepts, with dates: Jev model IDs and aliases by route. The Bedrock / Azure AI Foundry row was checked on 5 Oct, 05:17–05:21 UTC. Netlify's model IDs were read on 6, 7, 8 and 10 Oct (versions and aliases); its credit rate has not been re-read since 28 Sep, so the Netlify price keeps that date. No account login, charge or live request was made. Code for each route is on the Build pages. On 18 Oct 2026, 05:17–05:20 UTC the model IDs and aliases on all six routes were rechecked: only Opper changed (it now lists typesafe/jev-1.14.0); the other rows keep their 16 Oct values, unchanged on 18 Oct (R18-S05, R18-S22 to R18-S34; route table). On 23 Oct 2026, 05:19–05:23 UTC the model IDs and aliases were rechecked again: no listing change since 21 Oct, when Cloudflare's AI models catalog was first seen listing typesafe/jev-1.14.0; all six routes now list a 1.14 ID (listings and documented examples, not tested requests; R22-S05, R22-S22 to R22-S30; route table). Each route's image field: image input fields by route.

Change, 24 Oct 2026: the short answer and the TypeSafe row said jev-1.14.0 is a preview under jev-preview and jev-1.13.0 is under jev-latest (23 Oct 2026); since the move first seen on 24 Oct 2026, 05:12 UTC both aliases give jev-1.14.0 on TypeSafe's API and jev-1.13.0 is reached by its ID. "Before you send real data", the Netlify row and the Opper row add the 24 Oct alias check (R23-S05, R23-S27, R23-S30).

Change, 23 Oct 2026: the Cloudflare row said "no 1.14 ID listed at 18 Oct 2026, 05:20 UTC" (that check read the address without .0); it now names typesafe/jev-1.14.0. The TypeSafe row and "Before you send real data" said TypeSafe's alias "moves to 1.14 'in a week'" and, for Cloudflare, "you cannot pin"; they now give the 23 Oct alias state and limit the Cloudflare pin note to 1.13.

Resellers are not in this matrix: unofficial sites charge 3–16× the list price; see official vs reseller. Name collision: jev-router.com is an independent gateway for open models and does not serve TypeSafe's Jev; it is unrelated to OpenRouter's typesafe/jev-router.

Choosing on context and data terms

• 64k request budget: listed by TypeSafe direct and Opper, and in the text of Vercel's API description (whose context_window field says 32000). Cloudflare and Netlify list 32k. OpenRouter lists 262,144 since its 16 Oct 2026 entry, the size of TypeSafe's opt-in long-context limit (experimental, not production-ready per TypeSafe); it listed 64,000 / 32,000 from 10 Oct.
• Zero data retention: it must hold at both layers. For the TypeSafe provider, OpenRouter (ZDR list) and Cloudflare ("Yes") label it zero-retention and Vercel's docs mark it ✓, but Vercel also routes to DigitalOcean, which it marks as not ZDR. Opper says "not established". TypeSafe offers ZDR directly "for enterprise customers". Quotes and a checklist: data handling by route.
• No TypeSafe account: every gateway; OpenRouter states it explicitly.
• Exact cost per response: OpenRouter returns usage.cost with each response.

Third-party routes outside the six-route table

• Price and duration: OpenCode's docs: "Use jev-1.13-free instead of jev-1.13 to use the limited-time free model." No end date, quota, rate limit or request cap for it was found on that page (2 Nov 2026, 10:47 and 10:58 UTC; again 7 Nov 2026, 15:33 UTC, R27-S37). OpenCode's Terms: "If you are using a free version of the Services, we will notify you before any Services you are then using begin carrying a fee" (R26-S200, R26-S203).
• Data terms (OpenCode's own text, read 2 Nov 2026; the docs sentence, its exceptions list and the Terms' unpaid-account clause and effective date read again, unchanged, on 7 Nov 2026, 15:33 UTC, R27-S37, R27-S38): the docs say "Our providers follow a zero-retention policy and do not use your data for model training, with the following exceptions:", and the exceptions list has no Jev entry; the Terms (effective 15 Aug 2026) say "if you are using the Services through an unpaid account, we may use Content to further develop and improve our Services." and "we cannot and do not control the retention practices of Third Party Model providers." OpenCode does not say whether a request to the free model counts as an unpaid account, or which of the two texts controls. Both texts side by side: data handling by route.
• Which Jev version: OpenCode names "Jev 1.13" only, not a TypeSafe model ID or snapshot. No Jev ID was found in OpenCode's public model lists read without a key, in the models.dev registry, or in OpenCode's source read at commit ef6dd07 (2 Nov 2026, 10:47–10:49 UTC; R26-S201, R26-S202, R26-S205, R26-S207 to R26-S209).
• TypeSafe's side: TypeSafe's docs (135 pages via llms.txt), homepage and blog index did not mention OpenCode as a route on 2 Nov 2026, 10:49 UTC; the docs name opencode once, as a coding agent (R26-S215 to R26-S217, R26-S222).
• OpenRouter: none of the 14 IDs ending :free among 452 models on OpenRouter's model list is a Jev ID (2 Nov 2026, 10:47 UTC; R26-S218). On 7 Nov 2026, 15:29 UTC the list had 460 models and still no Jev ID ending :free (R27-S19).

Reported errors on this endpoint

Reported GitHub issues naming jev-1.13-free, all open when read on 2 Nov 2026, 10:48 UTC. They are third-party reports, not an uptime measure.

• can1357/oh-my-pi#15249 (opened 10 Oct 2026): the free variant missing from that tool's catalog (R26-S212).
• anomalyco/opencode#57451 (OpenCode's own repository, opened 20 Oct 2026): image requests to the endpoint "always fail with HTTP 400"; "Text-only requests through the same endpoint work fine." (R26-S210).
• DenisSud/pi-setup#3 (opened 21 Oct 2026): the free model returned 451 "Endpoint is unavailable" (R26-S211).
• apmantza/pi-free#637 (opened 25 Oct 2026): the ID was sent to a chat endpoint and returned 400 (R26-S213).
• dzhng/jevgrep#63 (opened 28 Oct 2026): "Model opencode/jev-1.13-free returns HTTP 400 "Model is unavailable"." (R26-S214).

Not tested by the pool: no OpenCode account or key was used and no request was sent to this route. OpenCode's Privacy Policy (effective 6 Mar 2026) has no free-model or Jev wording (searched 2 Nov 2026; R26-S204).

Is Jev on Amazon Bedrock or Azure AI Foundry?

Boundaries: AWS Marketplace and Azure Marketplace results were read on the first page only. Private offers, listings visible only when signed in, regional catalogs and enterprise contracts were not checked. This does not mean Jev will never be on either platform.

How to reach Jev from AWS or Azure today

These are the routes already in the route matrix; nothing here is new about them.

1. TypeSafe API directly (api.typesafe.ai/v1/systemone, key from the console). Prompts leave your VPC or tenant over the public internet to TypeSafe, which processes them on US compute (trust center). No PrivateLink or VNet option was found in TypeSafe's docs (boundary: llms-full.txt, 5 Oct 2026, 05:17 UTC).
2. OpenRouter (typesafe/jev-1.13; no TypeSafe account needed). Data goes to OpenRouter, then TypeSafe. The endpoint is on OpenRouter's ZDR list.
3. Vercel AI Gateway, Cloudflare, Opper or Netlify. Each adds a gateway hop before the provider. Opper is an EU-based gateway, but its TypeSafe route is listed as "United States".
4. From a Foundry agent: an OpenAPI tool that points at TypeSafe's API (the Microsoft Tech Community post below does this). The decision runs at TypeSafe, not in Azure.

What leaves the cloud

1. Your code on AWS or AzureLambda, AgentCore, a container, a Foundry agent
2. Public HTTPSoutside the Bedrock or Foundry model boundary
3. Gatewayoptional: OpenRouter, Vercel, Cloudflare, Opper, Netlify
4. TypeSafeUS compute: AWS, Modal, Nebius or CoreWeave (trust center)

Whichever route you use, the data path leaves AWS Bedrock's or Azure's model boundary, so Bedrock Guardrails and Foundry content filters do not see the Jev call. Retention, training and ZDR per route: data handling by route.

What others say

• Reported FactualMinds, a consultancy, in "Jev (TypeSafe AI) on AWS with Bedrock (2026)" (about 20 Sep): "Jev is not an Amazon Bedrock model. There is no inference profile, no VPC endpoint, and no Guardrails hop." It describes Jev as a sidecar called from Lambda or AgentCore over public egress to api.typesafe.ai.
• Reported Microsoft Tech Community, "Using Jev with Agents in Microsoft Foundry for Model Evaluation" (community blog, about 26 Sep): it builds a Foundry agent with an OpenAPI tool whose operation calls POST https://api.typesafe.ai/v1/systemone with a TypeSafe key. Jev is not a Foundry catalog model there.
• A lookalike reseller site, jevtypesafeai.com, has a "Jev on AWS Bedrock" page. It is not a source for this section; see official vs reseller sites.

Read on 5 Oct 2026: FactualMinds 05:20:20 UTC, Microsoft Tech Community 05:20:17 UTC. Query wording: "jev on bedrock" (Google), "jev bedrock" (Bing, DuckDuckGo) and "jev azure foundry" (Google, Bing, DuckDuckGo) were suggested at 05:15 UTC; a suggestion is a demand signal, not a volume.

Before you send real data

1. Pin jev-1.13.0 (or jev-1.14.0) instead of an alias if your thresholds depend on one model version. TypeSafe said on 14 Oct 2026 that jev-latest would move to jev-1.14.0 "in a week"; on 23 Oct 2026, 05:45 UTC, it still pointed to jev-1.13.0, and TypeSafe had published no date for the move. On 24 Oct 2026, 05:12 UTC, TypeSafe's models page listed jev-latest → jev-1.14.0; TypeSafe published no dated announcement of the move, and jev-1.13.0 is still listed and callable by its ID, with no deprecation date (R23-S05); other routes' aliases differ (jev-latest by route). On Cloudflare you cannot pin 1.13 (its 1.14 page shows requests that write typesafe/jev-1.14.0); Vercel lists versioned IDs since 16 Oct 2026; see what to pin on each route. Recheck aliases and limits before a pilot.
2. Use synthetic or redacted inputs until you have checked the account's access, retention, credits and terms. A public listing is not a successful call.
3. Read the TypeSafe MCA, privacy policy and DPA for your workload. On 10 Oct 2026 (05:12 UTC) the MCA still showed "last updated 23 Sep" (what changed from 19 Sep is unknown), the privacy policy 19 Nov 2025 and the DPA 24 Apr 2026. The training, retention and ZDR clauses are quoted on data handling by route (a reading of the text, not legal advice). Until its 19 Sep version the MCA also barred publishing benchmarks; see the terms box on the benchmarks page.
4. Handle the documented 401, 422, 429 and 529 responses with a fallback in your own code; see errors and rate limits.
5. Do not copy a context limit from one route to another: TypeSafe's 64k budget is not established for gateways that list 32k.