Thinking Machines Lab, the Mira Murati-founded AI company, launched Inkling 2 Small on 5 November 2026: a Mixture-of-Experts model with 594 billion total parameters, 17 billion active, image input, variable effort and a 1-million-token context window. It is available now on Tinker with closed weights; Apache 2.0 open weights are intended next week after external safety testing.

The launch continues the lab’s path since Mira Murati launched Thinking Machines Lab, and it lands as Indian enterprises weigh open-weight and agentic options alongside Indian open-weight and decision-model moves such as Sarvam Prism’s decision model.

What Inkling 2 Small delivers

On the Artificial Analysis Intelligence Index the model scores about 30.6 in the company’s benchmark table (company prose cites 31), well above prior Inkling generations. Coding marks include SWE-Bench Pro v2 Hard at 86.3%, DeepSWE v1.1 at 72.1% and FrontierCode v2.0 at 25.4%. Thinking Machines says the model is strong in agentic coding, vision, knowledge work and enterprise or finance benchmarks, with alignment improved versus earlier Inkling releases though still trailing some closed models.

Capabilities demos emphasise long-horizon autonomous work: iterative website and 3D design with visual feedback, coding across agent harnesses, and knowledge-work tasks that call tools over MCP. For CIOs comparing agent governance, that sits next to caution in the TCS enterprise AI study on how few firms let agents act alone.

How Thinking Machines trained the MoE

Pretrain phase H1 covered 8.4 trillion tokens in 68 hours on NVIDIA GB300 NVL72 racks, running natively in NVFP4. H2 added 3.8 trillion tokens of higher-quality data; H3 about 320 billion tokens of reasoning and agent trajectories at 1M context; H4 about 34 billion tokens of high-quality agent trajectories. A joint reinforcement-learning run then ran 1,075 steps with more than 17 million rollouts over 9 days on 2,048 GPUs, processing over 3 trillion tokens and more than 4×10²³ FLOPs.

Why open weights matter for enterprise buyers

Inkling 2 Small is live on Tinker for inference and fine-tuning with closed weights initially, following Thinking Machines’ stated safe path to open weights. Full weights under Apache 2.0 are intended next week after external safety testing and red-teaming. That timeline will interest Indian enterprise buyers and AI infrastructure builders watching open-weight stacks, and it pairs with scale debates such as GitHub’s view that AI agents could outnumber developers and governance tooling like Salesforce Agent Fabric’s kill switch and live evals.

Full details, benchmarks and limitations are in the Thinking Machines Lab Inkling 2 Small announcement. A model card is also published on the company’s site.

Key takeaways

• On 5 November 2026 Thinking Machines Lab launched Inkling 2 Small: 594B total / 17B active MoE, image input, variable effort, 1M-token context.
• Artificial Analysis Intelligence Index ~30.6 (table) / ~31 (company prose); coding scores include SWE-Bench Pro v2 Hard 86.3% and DeepSWE v1.1 72.1%.
• Available on Tinker now with closed weights; Apache 2.0 weights intended next week after external safety testing and red-teaming.

FAQ

What is Inkling 2 Small?

It is Thinking Machines Lab’s second-generation Mixture-of-Experts model: 594 billion total parameters with 17 billion active, supporting image inputs, variable effort and a 1-million-token context window.

Where can enterprises try Inkling 2 Small?

It is available now on Tinker for inference and fine-tuning with closed weights. Thinking Machines intends to release full weights under Apache 2.0 next week after external safety testing.

How does it compare on coding benchmarks?

Company-reported scores include SWE-Bench Pro v2 Hard at 86.3%, DeepSWE v1.1 at 72.1% and FrontierCode v2.0 at 25.4%, with an Artificial Analysis Intelligence Index near 30.6–31.

Featured image: Thinking Machines Lab Inkling 2 Small launch cover art — abstract black organic outlines over a dotted white grid. Photo: Thinking Machines Lab

Be a part of Elets Collaborative Initiatives. Join Us for Upcoming Events and explore business opportunities. Like us on Facebook , connect with us on LinkedIn and follow us on Twitter.

"Exciting news! Elets technomedia is now on WhatsApp Channels Subscribe today by clicking the link and stay updated with the latest insights!" Click here!