Inkling
#3 in Reasoning ModelsUnknown · 10× · last seen Jul 17, 2026
75
Momentum
Inkling is the first independently trained-from-scratch AI model from Thinking Machines Lab (founded by ex-OpenAI CTO Mira Murati), released on July 15, 2026 as an Open-Weights model under Apache 2.0 license. It is a Mixture-of-Experts Transformer with 975 billion total parameters (41 billion active) that natively operates multimodal across text, image, and audio, and supports a context window of up to 1 million Token. The model positions itself as an efficient, adaptable Reasoning generalist with controllable "Thinking Effort," but trails leading Chinese Open-Weights models such as GLM-5.2, Kimi K2.6, and DeepSeek v4 in independent Benchmarks.
Momentum trend
19.04.18.07.
Features
| __attribution_qa_at | 2026-07-18T05:30:40.702Z |
| Key Benchmark (%) | Artificial Analysis Intelligence Index: 41 (leading US open-weight model); SWE-bench Verified: 77.6% |
| Context Window (Tokens) | up to 1,000,000 tokens (open weights); 64K/256K on Tinker API |
| License | Apache 2.0 (plus separate Acceptable Use Policy) |
| Multimodality | Natively multimodal: text, image, and audio input, text output |
| Platform | Tinker (Thinking Machines), Hugging Face (weights), API via TogetherAI, Fireworks, Modal, Databricks, Baseten |
| Price per 1M Tokens | $1.87 input / $4.68 output (64K context, Tinker API, limited-time 50% discount) |
| Release Date | July 15, 2026 |