Published
Updated
TypeSafe-Danger-Zone-Long-Context raises both Jev context caps to 262k tokens
The header TypeSafe-Danger-Zone-Long-Context: true raises both context caps to 262k tokens on Jev 1.14 and Jev 1.13. Default limits stay 64k per request and 32k per question. On October 15, 2026, at 01:22 UTC, this desk's header calls returned HTTP 200 on 2,122 input tokens.
Diogo Almeida posted on October 14, 2026, at 21:52 UTC, that Jev already supported a lot longer contexts than TypeSafe had allowed. The tweet says that was a product choice, because the best uses likely should not need the extra room, and that context lengths should be up to 262k. The header is TypeSafe-Danger-Zone-Long-Context: true.
TypeSafe’s recap at 01:02 UTC on October 15 says Jev can take 4 times as much of it. 262,144 divided by 64,000 is 4.096. The link is the models page, context length.
Opened October 15, 2026, that models page still prints 64k per request and 32k per question on both Jev 1.14 and Jev 1.13. Those two limits are per_request_context, the state plus all questions, and per_question_context, the state plus any one question. There is no separate cap on how many questions a request may hold.
Long context is off by default. It works on both Jev 1.14 and Jev 1.13. With the header, both limits are 262k tokens. Requests without the header keep the table defaults. The page says long context costs the same per token, and that it is slower and less accurate than staying inside the defaults.
The blog says TypeSafe limited context on purpose because Jev, like all models, is less reliable on longer inputs, and because the constraint pushes callers toward smaller state.
The Jev Day index is Jev Day 2026. The model is jev-1.14.0. Images stay on their own header, TypeSafe-Danger-Zone-Images.
On October 15, 2026, the Date header 01:22:16 GMT, a POST to https://api.typesafe.ai/v1/systemone with model jev-preview and the long-context header returned HTTP 200 and named jev-1.14.0. The state is a termination-clause paragraph repeated until usage printed 2,122 input tokens. The Noul “Does the state include a clause that lets either party end the agreement early?” returned 0.95. Output tokens 25. The desk clock was 656 ms. List-price arithmetic at $0.042 per million input tokens is $0.000089.
The next second, Date header 01:22:17 GMT, the same state and header with model jev-latest returned HTTP 200 and named jev-1.13.0. That Noul was 0.96. Token counts matched. The desk clock was 448 ms. This desk did not fill 262k tokens.
Date headers 06:16:43 GMT and 06:16:44 GMT, two more header calls on a longer copy of that clause returned HTTP 200. jev-1.14.0 scored 0.96. jev-1.13.0 scored 0.95. Usage was 2,041 input tokens and 21 output tokens on both. This desk still did not fill 262k tokens.
Kerry Lu and Mateo Wang’s LiteLLM post, dated October 15, forwards TypeSafe-Danger-Zone-Long-Context: true on both decision routes, to TypeSafe only. This desk did not run the proxy. The write-up is LiteLLM.
This site's reading
Editorial notes evaluating claims against primary sources, contextualizing findings alongside related implementations, and defining technical terms.
Verify
Primary post is Diogo Almeida, @CompleteSkeptic, October 14, 2026, 21:52 UTC, status 2110488928906322286. The text says Jev already supported much longer contexts than TypeSafe allowed, that the cap was a product choice, that the best use cases likely should not need it, and that context lengths should be up to 262k. Header: TypeSafe-Danger-Zone-Long-Context: true.
TypeSafe's recap, 01:02 UTC October 15, status 2110536691702452264, says Jev can take 4 times as much context, and links docs.typesafe.ai/models#context-length. 262,144 divided by 64,000 is 4.096.
Opened October 15, 2026, the models page default table still prints 64k per-request and 32k per-question for Jev 1.14 and Jev 1.13. Long context is off by default and works on both. With the header, both limits are 262k tokens. Requests without the header keep the table defaults. Long context costs the same per token. The page says it is slower and less accurate than staying inside the defaults, so turn it on only for requests that need it. There is no separate limit on the number of questions. A request has to fit both the per-request cap and the per-question cap.
The blog says TypeSafe limited context on purpose because Jev, like all models, is less reliable on longer inputs, and because the constraint pushes callers toward smaller state.
On October 15, 2026, Date header 01:22:16 GMT, a POST to https://api.typesafe.ai/v1/systemone, model jev-preview, header TypeSafe-Danger-Zone-Long-Context: true, returned HTTP 200 and named jev-1.14.0. The state is a termination-clause paragraph repeated until the request reached 2,122 input tokens. Noul Does the state include a clause that lets either party end the agreement early returned 0.95. Output tokens 25. Desk clock 656 ms. List-price arithmetic $0.000089.
Date header 01:22:17 GMT, the same state and header, model jev-latest, HTTP 200, named jev-1.13.0. Noul 0.96. Same token counts. Desk clock 448 ms. List price $0.000089. This desk did not fill 262k tokens. Key, request JSON, and request ids are not recorded here.
Date headers 06:16:43 GMT and 06:16:44 GMT, two more header calls on a longer termination-clause state returned HTTP 200. jev-1.14.0 Noul 0.96. jev-1.13.0 Noul 0.95. Usage 2,041 input tokens and 21 output tokens on both. Desk clocks 783 ms and 427 ms. This desk still did not fill 262k tokens.
Kerry Lu and Mateo Wang's LiteLLM post, dated October 15, 2026, forwards TypeSafe-Danger-Zone-Long-Context: true on POST /v1/systemone and POST /v1/decisions, to TypeSafe deployments only. This desk did not run the proxy and did not fill 262k tokens on October 16.
Compare
Default TypeSafe limits on Access and FAQ, after the October 7 models-page reading, stay 64k for the whole request and 32k for state plus the longest question. Those sentences remain the no-header caps. 262k is the opt-in pair.
jev-doc-search walks a PageIndex tree when a flat Choice would pass 32k tokens or 255 options. That page's NVIDIA 10-K is given as about 76k tokens. Eikos's 64k hidden-decision row left Jev's cell blank because the API limit was 32k. Those two pages were written before this header.
2,122 input tokens is inside the default 32k cap. The October 15 calls show the header is accepted. They are not a 262k accuracy test. 656 ms and 448 ms are this desk's clocks on that short state.
Terms
- TypeSafe-Danger-Zone-Long-Context
- HTTP header, value true. Opened October 15, 2026, the models page sets both the per-request context and the per-question context to 262k tokens when this header is present. It works on Jev 1.14 and Jev 1.13.
- 262k
- 262,144 tokens. Diogo Almeida's October 14, 2026 tweet says context lengths should be up to 262k. TypeSafe's recap calls that 4 times as much as the default 64k per-request cap. 262,144 / 64,000 is 4.096.