What Happened

Thinking Machines Lab released Inkling on July 15, 2026—a 975B-parameter multimodal AI model under Apache 2.0, allowing free commercial use. It was one of nine open-weight models launched in 12 days, including releases from Moonshot AI, Zhipu AI, and Mistral AI. Inkling uniquely debuted alongside closed models like GPT-5.6 and Grok 4.5, highlighting the widening gap between self-hostable and API-dependent AI.

Who Is Affected

Enterprises weighing open vs. proprietary AI now have a US-built open alternative to Chinese models like GLM-5.2. Teams needing multimodal (text/image/audio/video) capabilities at scale are primary candidates, especially those sensitive to per-token API costs. Third-party inference providers can also integrate Inkling’s weights.

Market Impact

• Cost pressure: Inkling’s 50% launch discount on Tinker hosting undercuts closed-model APIs.
• Benchmark shift: Scored 78.0% on FORTRESS Adversarial, leading open models.
• Timing: Surrounded by four closed releases (GPT-5.6, Grok 4.5, Muse Spark 1.1), making open/closed tradeoffs tangible.

What to Watch

Adoption of Inkling-Small (released July 31) for cost-sensitive deployments. Response from Chinese labs, which delivered 5/9 confirmed July open releases. Whether Thinking Machines’ quiet build approach becomes a template for future challengers.

Related coverage

• MiniMax H3 releases multimodal video generation model amid copyright lawsuit
• SpaceXAI releases Grok 4.5, described as 'Opus-class model'
• Giotto.ai Launches Free Cloud Access to Its AI System
• Anthropic launches Claude Science AI research workbench