Open sourced typed inference for software (SN100).
Bittensor
Joined September 2026
- Tweets11
- Following2
- Followers34
- Likes3
Ours is already live on Bittensor: Neon 1.1 routes every task across 32 models by difficulty, and miners on Subnet 100 make it better every day, try it at console.opentype.dev
You choose which models the router can use: if you already have a Codex or Claude Code subscription, we recommend filtering to GPT-6 Sol and Claude Opus 5.5, the best models in our tests, with GPT-6 Astra as an option at a slightly higher cost. GLM-5.3 is also a great choice, and DeepSeek V4.1 Flash is the pick for execution speed.
Meet Neon 1.1 by OpenType.
A model built to decide, not to chat: hundreds of questions answered in one call, in milliseconds.
We ran it through 5 real tests. Every number in this thread comes from a logged run. 🧵
Every number in this thread comes from Neon 1.1 served as is: zero serving optimisations so far.
The endpoint scales automatically with traffic, and optimisations land over the coming weeks to push latency down as far as it goes.
Try Neon 1.1 today: $5 in free credits when you sign up. → opentype.dev
