Uploaded September 2026 | Updated November 2026, 3 hours ago
Jev came out September 15th, 2026, and it doesn't work like a normal frontier model. No chat, no text back. You send it state plus a list of yes-or-no questions and it sends back probabilities, in 70 to 500 milliseconds, at $42 per billion input tokens with output free.
Check out my last video on if the frontend is dead!
youtube.com/watch?v=ucJMazXR3X8
I built two apps to see if that's useful. Stack Gate runs five semantic checks on a real Vue.js Skills pull request, and I compare OpenRouter against TypeSafe's API directly (683ms vs 318ms). Jev Cost Advisor classifies a workload description, then deterministic MCP servers hit AWS pricing to produce the actual estimate. Plus the community showcase, including Super Mario Brothers running on Jev calls.
Traditional LLMs aren't going away, but if you need thousands of small decisions feeding a deterministic flow, this is worth a look. Let me know in the comments if you'd use it.
๐ Links
- Jev announcement: typesafe.ai/blog/introducing-system-one-models-and-jev
- TypeSafe docs: docs.typesafe.ai/introduction
- Confidence and calibration: docs.typesafe.ai/confidence
- Jev on OpenRouter: openrouter.ai/~typesafe/jev-latest
- TypeSafe evals: evals.typesafe.ai
- Agent Toolkit for AWS: github.com/aws/agent-toolkit-for-aws
- AWS MCP Server docs: docs.aws.amazon.com/agent-toolkit/latest/userguide/mcp-server.html
- AWS Pricing MCP server: github.com/awslabs/mcp
Follow me on socials!
X: https://x.com/ErikCH
LinkedIn: linkedin.com/in/erikhanchett
Bluesky : https://bsky.app/profile/erikch.bsky.social
Website: programwitherik.com
https://www.kiro.dev - my favorite harness!
Join my newsletter and get updated when I release cool stuff - programwitherik.com/newsletter
0:00 What is Jev?!
01:03 noul questions app Stack Gate
03:48 Jev cost advisor explained
05:33 Other cool Jev examples
Jev came out September 15th, 2026, and it doesn't work like a normal frontier model. No chat, no text back. You send it state plus a list of yes-or-no questions and it sends back probabilities, in 70 to 500 milliseconds, at $42 per billion input tokens with output free.
Check out my last video on if the frontend is dead!
youtube.com/watch?v=ucJMazXR3X8
I built two apps to see if that's useful. Stack Gate runs five semantic checks on a real Vue.js Skills pull request, and I compare OpenRouter against TypeSafe's API directly (683ms vs 318ms). Jev Cost Advisor classifies a workload description, then deterministic MCP servers hit AWS pricing to produce the actual estimate. Plus the community showcase, including Super Mario Brothers running on Jev calls.
Traditional LLMs aren't going away, but if you need thousands of small decisions feeding a deterministic flow, this is worth a look. Let me know in the comments if you'd use it.
๐ Links
- Jev announcement: typesafe.ai/blog/introducing-system-one-models-and-jev
- TypeSafe docs: docs.typesafe.ai/introduction
- Confidence and calibration: docs.typesafe.ai/confidence
- Jev on OpenRouter: openrouter.ai/~typesafe/jev-latest
- TypeSafe evals: evals.typesafe.ai
- Agent Toolkit for AWS: github.com/aws/agent-toolkit-for-aws
- AWS MCP Server docs: docs.aws.amazon.com/agent-toolkit/latest/userguide/mcp-server.html
- AWS Pricing MCP server: github.com/awslabs/mcp
Follow me on socials!
X: https://x.com/ErikCH
LinkedIn: linkedin.com/in/erikhanchett
Bluesky : https://bsky.app/profile/erikch.bsky.social
Website: programwitherik.com
https://www.kiro.dev - my favorite harness!
Join my newsletter and get updated when I release cool stuff - programwitherik.com/newsletter
0:00 What is Jev?!
01:03 noul questions app Stack Gate
03:48 Jev cost advisor explained
05:33 Other cool Jev examples










