Thinking Machines Lab has released the full weights for their Inkling 2 Small model under the permissive Apache 2.0 license. This allows users to download, inspect, fine-tune, and self-host the model for commercial use. While the model boasts a 1-million-token context window and scores 31 on the Artificial Analysis Intelligence Index, matching Gemini 3.8 Flash, self-hosting the 594-billion-parameter model requires significant hardware resources, with checkpoints needing 500-800GB of GPU memory. AI
IMPACT Enables broader adoption and commercial use of a high-performance, large-context model, though self-hosting requires substantial infrastructure.
RANK_REASON Open-weights release of a large, high-performing model by a named lab. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
- Apache Software License 2.0
- Artificial Analysis
- Artificial Analysis Intelligence Index
- Baseten
- Fireworks
- Gemini 3.8 Flash
- Hugging Face
- Inkling 2 Small
- Nebius Group
- Nvidia
- SGLang
- Snowflake Cortex Training
- Thinking Machines Lab
- Together
- TokenSpeed
- Unsloth
- vLLM
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →