Thinking Machines Lab releases Inkling, a 975-billion-parameter open-weight AI model
Open source release ● Confirmed 95% confidence first seen
Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, released Inkling, an open-weight foundation model with 975 billion parameters trained on audio, video, and text. The model is positioned as a cost-efficient alternative to closed models from major AI companies, requiring fewer tokens for comparable performance and offering customization through the Tinker fine-tuning platform.
Decision brief
- What changed
- Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, publicly released Inkling, a 975-billion-parameter open-weight foundation model trained on audio, video, and text, alongside its Tinker fine-tuning platform for customization.
- Why it matters
- This gives enterprises a downloadable, modifiable alternative to closed models from OpenAI, Anthropic, and Google, potentially lowering compute costs (per company benchmarks, it needs a third as many tokens as Nvidia's Nemotron 3 Ultra for equivalent coding performance) and reducing vendor lock-in. For technology and cost-sensitive leaders, it signals a viable path to self-hosted or fine-tuned AI infrastructure rather than dependence on API-based closed models, though real-world validation outside vendor benchmarks is still needed.
- Evidence
- Three independent outlets (TechCrunch, Wired, TLDR) consistently report the model's parameter count, open-weight nature, and positioning against closed AI giants; the token-efficiency comparison to Nemotron 3 Ultra and coding benchmark come specifically from TechCrunch citing the company's own benchmarks.
- What remains uncertain
- Performance and cost claims are based on Thinking Machines' own benchmarks rather than independent third-party testing, so real-world efficiency, output quality, and total cost of ownership versus closed models remain unverified. It's also unclear how enterprise adoption, licensing terms, and support compare to established open-weight competitors like Chinese open models mentioned in coverage.
- Monitor next
- Watch for independent benchmark testing or early enterprise adoption reports of Inkling to validate the claimed cost and token efficiency advantages.
Analytical support, not advice — assumptions and open questions stated above.