Jev Day 2026 shipped Jev 1.14.0, plus opt-in images, a Box type, and 262k context.

TypeSafe used Jev Day for the October 14, 2026 release. The dated stories on this desk are jev-1.14.0, images, Box, and 262k context. What a call returns is on what Jev is.

Story	One figure
jev-1.14.0	At 00:28 UTC October 16, jev-latest still returned jev-1.13.0. Yamashita Kyoto-ben 0.833 against 0.817. SosyalDijital 29 of 30 on both models.
Images	Continue button noul 0.99 on 1.14.0, n = 3. Empty field 0.02. Jev4Mellea 12 scenes, coverage 5 of 12 at 0.10 / 0.90.
Box	Aviz 82% on 180 scenes, 0.21 s, $0.15 per 1k, ≥0.7 right 107 of 108. Continue n = 3 mean IoU 0.93, confidence 0.91. HTTP 400 on jev-1.13.0.
Long context	Both caps 262k with TypeSafe-Danger-Zone-Long-Context.

What is Jev Day 2026?

Jev Day 2026 is TypeSafe's October 14, 2026 release of Jev 1.14.0, plus opt-in images, a Box type, and 262k context.

TypeSafe posted a teaser at 17:16 UTC, what's this, Jev is evolving, and a town hall with Diogo Almeida and Allie at 1:14 PM. Almeida's reply named 1:14 PM PST and Discord. 1:14 PM PDT on October 14 is 20:14 UTC. His Happy JevDay thread is 21:52 UTC. The blog is dated October 14. TypeSafe's recap thread is 01:02 UTC on October 15, and says the company announced Jev 1.14 on 10.14 at 1:14pm PT.

What is jev-1.14.0?

jev-1.14.0 is TypeSafe's October 14, 2026 Jev release, shipped as jev-preview.

The models page, opened October 15, prices Jev 1.14 and Jev 1.13 at $0.042 per million input tokens, output free, 64k per request and 32k per question. Both rows still print Input Text. The alias table names jev-latest as jev-1.13.0 and jev-preview as jev-1.14.0. The blog says jev-latest will move to 1.14.0 in a week, that 1.13.0 stays available, and that TypeSafe intends 60 days of warning before 1.13.0 is formally deprecated, and 30 days for 1.14.0. The dated write-up is jev-1.14.0.

On October 16, 2026, at 08:19 UTC, one Box for every vehicle returned one rectangle, n = 3, and type array returned HTTP 400. Those numbers sit on Box.

On October 15, 2026, Date header 01:22:12 GMT, GET https://api.typesafe.ai/v1/models listed those two aliases. Date header 01:22:13 GMT, a POST with model jev-latest returned HTTP 200 and named jev-1.13.0. A POST with model jev-preview the same second named jev-1.14.0.

Later that UTC day, Date headers 03:03:53 GMT to 03:03:56 GMT, six POSTs with model jev-preview named jev-1.14.0. Four payouts tickets in English, Chinese, Spanish, and Japanese came back at language confidence 1.0, all billing. Chinese frustration scored 1.37 at confidence 0.44. A Chinese API-500 ticket chose technical at 1.0. Matching jev-latest calls named jev-1.13.0 and also returned language 1.0 on those four sentences. The dated numbers sit on jev-1.14.0.

OpenRouter posted at 00:16 UTC that Jev 1.14 is now on OpenRouter. Opened later that UTC day, typesafe/jev-1.14 lists $0.042 per million input tokens, output free, and a 262K context. ~typesafe/jev-latest already linked Jev 1.14. GET /api/v1/models did not list that id. This desk did not call OpenRouter.

Date headers 06:16:20 GMT to 06:16:44 GMT, 32 POSTs plus GET /v1/models. TypeSafe's jev-latest still named jev-1.13.0. Three Chinese payouts tickets on jev-1.14.0 scored frustration 1.27, 1.32, and 1.32, mean 1.30 at mean confidence 0.54. Matching jev-1.13.0 mean 1.54 at 0.31. Language was 1.0 on seven languages on both models.

Date header 11:04:52 GMT, GET /v1/models still listed those two aliases. A POST with model jev-latest still named jev-1.13.0. A POST with model jev-preview named jev-1.14.0. Opened that hour, typesafe/jev-1.14 still returned HTTP 200. GET /api/v1/models still omitted the id.

Date headers 14:57:05 GMT to 14:57:28 GMT, GET /v1/models still listed those two aliases. A POST with model jev-latest still named jev-1.13.0, noul 0.89, billing 1.0 at confidence 1.0. A POST with model jev-preview named jev-1.14.0, noul 0.90, billing 0.99 at 0.98. Opened that hour, typesafe/jev-1.14 still returned HTTP 200. GET /api/v1/models still omitted the id. The dated numbers sit on jev-1.14.0.

Date header 00:28:46 GMT October 16, 2026, GET /v1/models returned HTTP 200. Date header 00:29:34 GMT, a second GET named jev-latest and jev-preview, both release_date still 2026-09-10. A POST with model jev-latest named jev-1.13.0, noul 0.89, billing 0.87 at 0.80. A POST with model jev-preview named jev-1.14.0, noul 0.90, billing 0.86 at 0.79. Opened that hour, typesafe/jev-1.14 listed $0.042, 262K, and TypeSafe P50 0.19 s. GET /api/v1/models still omitted the id. This desk did not call OpenRouter. PC Watch reprinted the blog. LiteLLM added day-0 images, box, and long context. The dated numbers sit on jev-1.14.0.

What is the danger zone?

Danger zone is TypeSafe's name for opt-in headers on features it will not yet vouch for in production.

Images use TypeSafe-Danger-Zone-Images: true. Long context uses TypeSafe-Danger-Zone-Long-Context: true. The blog says those features will stay on later models, that they still have too much jaggedness for production, and that Box is only on Jev 1.14. Almeida also posted a console Organization toggle, Help improve TypeSafe models, that lets TypeSafe read inputs and outputs to find jaggedness. The blog says TypeSafe still will not train on those queries. This desk did not turn that toggle on.

How do images work on Jev?

An image sits in the assets map and can be referenced from state, instructions, or criteria.

The images docs, opened October 15, require the images header on every request that contains an image or a Box question. Jev 1.13 and Jev 1.14 both accept images. Formats are PNG or JPEG. Each side is 256 to 1568 pixels. A 256 by 256 square is too small. The dated write-up is images.

On October 15, Date header 01:22:14 GMT, a POST with an image and no header returned HTTP 400, Images are experimental. With the header, a 400 by 400 red-square PNG scored noul 0.90 on jev-1.13.0 and 0.39 on jev-1.14.0.

Date headers 06:16:36 GMT to 06:16:40 GMT, three more present squares scored 0.49, 0.47, and 0.42 on jev-1.14.0, mean 0.46. Matching jev-1.13.0 Nouls scored 0.94, 0.94, and 0.93. An empty field scored 0.04 and 0.01.

Date headers 11:04:54 GMT to 11:04:57 GMT, an 850 by 850 PNG in state used 2,238 input tokens on jev-1.14.0. Two questions that each named that file in instructions used 4,209 tokens. Wes Bos billed $0.000096936 for one 850 by 850 question.

Jev4Mellea's twelve-scene capture, started 11:45:44 UTC on jev-1.14.0, accepted none of 6 positives at 0.10 / 0.90, coverage 5 of 12. Date headers 14:57:24 GMT to 14:57:27 GMT, a two-square PNG scored blue noul 0.98 and red noul 0.56 on jev-1.14.0, n = 3. Matching jev-1.13.0 red mean 0.87. The dated numbers sit on images.

Date headers 00:28:47 GMT to 00:28:48 GMT October 16, a Checkout PNG scored Continue noul 0.98, 0.99, 0.99 on jev-1.14.0, mean 0.99. Matching jev-1.13.0 scored 1.0. An empty field scored 0.02. Four-color Nouls scored 1.0. A red circle among eighty gray circles scored 0.99, 0.99, 0.98.

What is a Box?

A Box answer is one rectangle. On October 16, 2026, at 08:19 UTC, one Box for every vehicle returned one rectangle, n = 3, confidence 0.19, 0.17, 0.13, and type array returned HTTP 400.

TypeLLM's cookbook at 14.00 s prints six vehicle boxes, including Tesla's flying Roadster. Named Boxes in one request on this desk's three-car PNG scored IoU 0.89, 0.89, 0.90. A 960 by 540 frame from Elon Musk's clip chose six at 0.59, in-air noul 0.94. Continue-button n = 3 averaged IoU 0.93.

The Box docs say there is no not-found value, so a missing target still returns four edges. The page tells the caller to ask a Noul whether the target is present. Box takes no criteria. The dated write-up is Box.

On October 15, Date header 01:22:15 GMT, this desk's box around a red square drawn at 0.40 to 0.80 and 0.20 to 0.70 came back left 0.4, right 0.8, top 0.16, bottom 0.7, confidence 0.93, on jev-1.14.0. The Noul on that request was 0.39. The same Box question on jev-1.13.0 returned HTTP 400.

Date headers 06:16:36 GMT to 06:16:40 GMT, three more boxes on that PNG averaged left 0.40, right 0.79, top 0.16, bottom 0.71, confidence 0.91. An empty field still returned a box, at confidence 0.05.

Date headers 14:57:24 GMT to 14:57:25 GMT, three boxes on a blue square drawn at 0.08 to 0.28, with a red square beside it, averaged left 0.07, right 0.30, top 0.14, bottom 0.29, confidence 0.94. Date header 14:57:28 GMT, the same Box on jev-1.13.0 returned HTTP 400.

Date headers 00:28:47 GMT to 00:28:48 GMT October 16, three boxes on a green Continue button drawn at 0.55 to 0.90 and 0.70 to 0.88 averaged left 0.55, right 0.91, top 0.71, bottom 0.88, confidence 0.91, mean IoU 0.93. Date header 00:28:49 GMT, the same Box on jev-1.13.0 returned HTTP 400. Four colors in one request used 4,458 input tokens against 6,612 as four separate POSTs.

How long is long context?

With TypeSafe-Danger-Zone-Long-Context: true, both context caps are 262k tokens on Jev 1.14 and Jev 1.13.

Default limits stay 64k per request and 32k per question. TypeSafe's recap calls 262k four times the 64k cap. 262,144 / 64,000 is 4.096. The models page says long context costs the same per token, and that it is slower and less accurate than the defaults. The dated write-up is long context.

On October 15, Date headers 01:22:16 GMT and 01:22:17 GMT, this desk's header calls on 2,122 input tokens returned HTTP 200 on jev-1.14.0 and jev-1.13.0. This desk did not fill 262k tokens.

What did SosyalDijital score?

SosyalDijital scored both Jev 1.13 and Jev 1.14 at 29 of 30 on 30 synthetic Turkish tickets, and both marked the same buyer as a seller.

The article, opened October 15, used one pass of 30 synthetic messages, labels buyer, seller, and other, 60 API calls. The shared miss is "Reklam değil gerçek talep: ürün fotoğraflarını düzenleteceğim, ücret konuşalım." Expected buyer, both returned seller. Recorded model fields typesafe/jev-1.13-20260917 and typesafe/jev-1.14-20261014. At 11:04 UTC, this desk's Choice on that sentence chose seller on all three jev-1.14.0 calls, at confidence 0.48, 0.43, and 0.52. Matching jev-1.13.0 calls chose seller, buyer at 0.28, then seller. The dated write-up is jev-1.14.0.

What did Yamashita score on Kyoto-ben?

Yamashita scored jev-1.14.0 at 0.833 and jev-1.13.0 at 0.817 on 60 Kyoto-ben tickets, three-run means, and majority vote matched on all 60.

The Qiita write-up, dated October 15, used 60 frozen phrases, six labels, 420 calls, about $0.02. McNemar p = 1.000. One majority change sat in ten later phrases: "うちの子ぉも、ええ子にしてますえ", expected warning. At 14:57 UTC, this desk's Choice on that sentence, criteria written here, chose social_nicety on all three jev-1.14.0 calls, at confidence 0.69, 0.66, and 0.64, and on all three jev-1.13.0 calls, at 0.33, 0.36, and 0.34. AI BENCHY, opened the same day, scored Jev 1.14 at 2.3 against Jev 1.13 at 2.6. The dated write-up is jev-1.14.0.

What did TypeSafe print for intelligence?

TypeSafe's internal chart, posted October 14, 2026, prints overall intelligence 60.0 for jev 1.14.0 and 58.9 for jev 1.13.0.

The image is watermarked antibench.com and says Please don't trust them. Calibration error on the next chart is 1.6 for 1.14.0 and 1.8 for 1.13.0, lower better. Almeida wrote that the measure is an experiment, likely biased toward TypeSafe's models. 60.0 is not JevBench's hosted composite. Opened October 11, 2026, that board ranks Jev 1.13.0 fourth at 71.5. The numbers sit on jev-1.14.0 and on Evals.

What did Aviz Maeir score?

Aviz Maeir scored Jev 1.14 Box at 82% on 180 scenes, 0.21 s, $0.15 per 1k, against gpt-6-luna 84% in 1.5 s and gpt-6.1-sol 95% in 3.4 s.

He posted at 22:04 UTC on October 15. Confidence 0.7 or above was right 107 of 108. Luna printed 0.9 to 1.0 on 27 confident misses. Where's Waldo on 30 to 120 people scored 70 of 72. This desk did not rerun the 180. Date headers 08:19:44 GMT to 08:19:45 GMT October 16, this desk's every-vehicle Box returned one rectangle, n = 3. The dated write-up is Box.