Inkling-Small: 276B sparse MoE, 12B active, Apache 2.0
Thinking Machines released Inkling-Small under Apache 2.0 — a 276B sparse mixture-of-experts model with 12B active parameters that scores 40 on the Artificial Analysis Intelligence Index, one point behind the 975B Inkling from two weeks ago. Near-frontier reasoning (95.5% AIME 2026, 80.2% SWE-bench Verified) now runs on open weights with 12B active parameters, and 25 quantized variants ship across llama.cpp, Ollama, LM Studio and Jan.
Source: huggingface.co ↗
Inkling-Small is a general-purpose multimodal model that accepts text, image and audio inputs and generates text outputs.
Thinking Machines
Why this matters
- → Frontier reasoning (95.5% AIME, 80.2% SWE-bench) now runs on 12B active parameters under open weights.
- → Apache 2.0 license enables research, fine-tuning, and third-party integration without restrictions.
- → 25 quantized variants ship across llama.cpp, Ollama, LM Studio, Jan for immediate local deployment.
Open frontier reasoning