KTwo AI Labs
Frontier, owned.
The first model that trains itself.
KTwo Crust 1 Beta is a 9B model, multimodal from the start, and the first in the world that trains itself natively. It is in private beta and not publicly available.
- Status
- Private beta
- Parameters
- 9B
- Modality
- Native multimodal
- Per 1M tokens
- $0.04 in, $0.15 out
Most models stop learning the day they ship. Crust 1 keeps going. It trains itself, natively, and no model did that before it. All in 9B parameters.
Models
Two models. Both train themselves.
Crust 1 Beta is in private beta. Mantle 0.5 isn’t publicly available yet.
LE PEAK
Twenty-three models. One thread.
LE PEAK is our chat. Twelve models are free, no account needed. Sign in for eleven more, each with its own daily quota, and switch between them mid-thread.
- NVIDIA NIM
- 8 models
- Step 5 Preview, DeepSeek V4.1 Flash, GLM 5.3 Flash, Muse Glimmer 30B, Laguna XS 2.1, Nemotron 3.5 Super VL, Nemotron 3.5 Lightning, Nemotron 3 Ultra Free, no account. 20 messages a minute.
- OpenRouter Free
- 4 models
- gpt-oss 120B Free, Gemma 4 26B A4B, MiniMax M3.1 Flash, North Mini Code Free, no account. 20 messages a minute.
- OpenRouter Premium
- 2 models
- GPT Luna, Claude Haiku Free account. 10 requests or 10,000 tokens a day, per model.
- Xiaomi MiMo
- 4 models
- MiMo V2.6 Pro, MiMo V2.6 Flash, MiMo V2.5 Pro, MiMo V2.5 Free account. 10 requests or 10,000 tokens a day, per model.
- Ollama Cloud
- 2 models
- Beam, gpt-oss 120B Free account. 40,000 tokens a day, per model.
- TypeSafe Jev
- 3 models
- Jev 1.14.0, Jev Preview, Jev Latest Answers with decisions, not text. Free account. 1,000,000 input tokens a day, per model; output is free.
- Attachments
- Every model
- Text files and PDFs on all 23; images, video (as frames) and scanned pages on the 12 Vision models and on Jev with Danger Zone Images; audio on MiMo. Prepared in your browser; nothing is stored.
Replies come from these third-party models, not from KTwo’s own: Crust 1 Beta and Mantle 0.5 aren’t in the chat. Don’t paste secrets. What the chat keeps about you is in Privacy.
Blog
Notes from the climb.
Self-training, multimodality, token pricing, Hungarian law and how LE PEAK is built. Every post.
-
Research
Mixture of experts, explained: why 120B can run like 5B
How mixture-of-experts AI models work, what names like 30B-A3B and 550B-A55B mean, and the trade-off between total and active parameters in memory, speed and cost. -
Guide
What is a token? Tokenization, and why Hungarian costs more
AI models read tokens, not words. How tokenization works, how to estimate token counts, and why the same text in Hungarian takes about twice the tokens of English. -
Guide
What is a context window? AI memory limits, explained
A model’s context window is how much text it can read at once. What it counts, why long contexts cost memory and money, and why bigger isn’t always better.
Access
Request access to Crust 1 Beta.
The beta is invite-only. Leave your work email and we’ll reply if a seat opens.
An inquiry or access request does not promise access to any AI model on this site. KTwo currently offers no public services.