Mira Murati’s Thinking Machines Lab Ships Inkling, an Open-Weights AI Model Aimed at Western Developers

Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, has released its first AI model — an open-weights system called Inkling — arriving nearly two years after she walked away from one of the most powerful seats in the industry.

Inkling is fully open-weights. Developers can access and deploy the underlying model weights directly rather than routing everything through a gated API. The Wall Street Journal frames the release as a bid to “loosen AI giants’ grip” on the market — positioning Murati’s venture as a counterweight to the closed-system dominance of OpenAI, Anthropic, and Google DeepMind.

The strategic logic isn’t complicated. Western developers hunting for open-weights models have largely turned to Chinese labs — DeepSeek, Moonshot AI, Alibaba’s Qwen team — because no comparably credentialed Western lab has been shipping freely available weights at scale. Inkling changes that calculus, at least on paper. According to Decrypt, the model gives Western developers something they haven’t had before: a serious open-weights option from a Western lab run by someone with genuine top-tier AI executive pedigree. Decrypt is also blunt about the ceiling, reporting that Inkling is not expected to dethrone the leading Chinese open-weights models on benchmarks.

That admission matters. The open-weights space is ferociously competitive, and Chinese labs have set a punishing bar. Moonshot AI recently released Kimi K3, a 2.8-trillion-parameter model that drew real attention from developers and researchers. DeepSeek’s V3 and R1 have similarly pushed what open-weights systems can do in reasoning and coding. Inkling enters this field carrying the weight of Murati’s name and her OpenAI lineage. Pedigree, though, doesn’t move benchmarks.

Murati served as Chief Technology Officer at OpenAI, overseeing the development and deployment of some of the most consequential AI systems the industry has produced. She departed in September 2024 — a move that sent ripples through the AI world at a moment when OpenAI was already dealing with leadership turbulence. The founding of Thinking Machines Lab followed. The roughly two-year runway from departure to first model release suggests a deliberate, build-first approach rather than a scramble to ship something half-finished for the headlines.

The technical specifics are still murky. Thinking Machines Lab has not publicly confirmed Inkling’s parameter count, architecture, or licensing terms. Social media has circulated a figure of 975 billion parameters in connection with Murati, but that number is unverified and does not appear in reporting from either Decrypt or the Wall Street Journal — treat it as unconfirmed until the company says otherwise. Without confirmed specs or independent benchmark results, there is no reliable way to assess how Inkling stacks up against DeepSeek V3, Kimi K3, or Meta’s Llama series on standard evaluations. The intended use case — whether Inkling is optimized for coding, general reasoning, or multimodal tasks — is also unspecified in available reporting.

Those gaps won’t stay open for long. The developer community pressure-tests new open-weights models within hours of release, and Inkling will face the same treatment. If it underperforms relative to the Chinese alternatives it is implicitly challenging, the narrative of a Western open-weights champion will be hard to sustain regardless of Murati’s reputation.

The stakes extend well beyond any single model. Inkling’s release is one data point in an escalating East-versus-West open-source AI race — one in which Chinese labs have moved aggressively while most Western players, Meta being the notable exception, have kept their most capable systems locked behind API gates. Murati’s bet is that genuine demand exists for a Western alternative, and that a startup rather than a trillion-dollar incumbent can fill it. Whether that bet pays off depends entirely on what Inkling can actually do once developers get their hands on it — and whether Thinking Machines Lab can iterate fast enough to keep pace with Chinese labs already working on their next release.