Thinking Machines, the AI startup founded by former OpenAI CTO Mira Murati, released Inkling, an open-source multimodal language model built for enterprises running on-premises or private cloud infrastructure. The company positioned the model as a cost-effective alternative to proprietary systems while emphasizing user control and customization.
Inkling launches under an Apache 2.0 license, which permits commercial use and modification without restrictions. The model demonstrates strong performance on software engineering benchmarks, achieving 77.6% on SWE-bench Verified, outperforming Nvidia Nemotron 3's 71% score. This positioning targets a specific market segment: organizations seeking open-weights models they can fine-tune, deploy locally, and operate without external dependencies.
The release reflects a broader shift in enterprise AI adoption. Companies increasingly want to avoid vendor lock-in and regulatory exposure that comes with relying on third-party APIs. Thinking Machines explicitly framed Inkling as addressing concerns about censorship and content filtering, appealing to organizations that need unfettered model access for their use cases.
The startup's emphasis on "resistance to censorship" distinguishes it from many competitors but also signals potential tension with safety-conscious deployment practices. The framing suggests the model minimizes built-in content restrictions, which could appeal to enterprises but also raises questions about moderation capabilities for harmful outputs.
Thinking Machines positions itself against both closed models from OpenAI and Anthropic and open competitors like Meta's Llama family. The sub-state-of-the-art performance claim acknowledges Inkling won't match frontier models on general benchmarks, but the engineering task specialization and on-premises deployment flexibility target practical enterprise needs rather than raw capability.
The open-source release represents a calculated business strategy. By licensing under Apache 2.0