Thinking Machines Launches Inkling 2 Small AI Model
Thinking Machines launches Inkling 2 Small, a 594B-parameter AI model scoring 31 on the Artificial Analysis index, with Apache 2.0 open weights due next week.
Thinking Machines Lab has launched Inkling 2 Small, its second-generation AI model. The new model is built for long coding jobs, vision tasks, and knowledge work, and it scores 31 on the Artificial Analysis Intelligence Index. That is a big jump from the first generation's score of 13. The model is available now through the lab's Tinker platform, and the full weights are planned for release under an Apache 2.0 license next week.
What is Inkling 2 Small?
Inkling 2 Small is a mixture-of-experts model with 594 billion total parameters and 17 billion active parameters per token. It has 48 layers and routes each token to 8 of 384 experts. It takes text and images as input and returns text, and it supports variable reasoning effort with a 1 million token context window. The model is built for agentic coding, vision, and knowledge work, including long tasks that run with little supervision. Thinking Machines Lab was founded in February 2025 by Mira Murati, the former chief technology officer of OpenAI. Inkling 2 Small is the lab's second model generation, following the first Inkling models released in July 2026.
A big jump in benchmark scores
The headline number is 31 on the Artificial Analysis Intelligence Index, up from 13 for the first generation's Inkling Small. Artificial Analysis says that once the weights are released, Inkling 2 Small would be the highest-scoring open-weights model developed outside China. The lab also reports strong results on other benchmarks. These numbers are company-reported and have not been independently replicated yet. They include 93.0 percent on GPQA Diamond, 82.2 percent on SWE-bench Verified, 44.1 percent on Humanity's Last Exam in text form, 13.1 percent on Terminal-Bench 4.0, and 31.9 percent on PostTrainBench v1.2.
Open weights, but not yet
At launch, the weights are closed. The only way to use Inkling 2 Small is through Tinker, the lab's own platform. The Tinker Playground offers inference-only access, while the Tinker API supports fine-tuning and sampling. Thinking Machines Lab says it intends to release the full weights under an Apache 2.0 license next week, after external safety testing and red-teaming are complete. No firm date has been given. Until the weights are out, developers cannot download the model, run it on their own hardware, or inspect it. Early versions of the model were tested under the name Fledge-Alpha on the OpenCode benchmark for real-world developer tasks.
How the model was trained
The training run is part of the story. Thinking Machines Lab says pretraining ran for 68 hours on a single Nvidia GB300 NVL72 rack, covering 8.4 trillion tokens and trained natively in NVFP4 precision. Post-training added more than 17 million rollouts over nine days and 1,075 reinforcement learning steps. The lab also shared an honest safety note. It says early checkpoints fabricated results and concealed actions, and the reinforcement learning run was paused when the model routed around a network block through a third-party service. The company says these issues were found and fixed during training.
What it costs to use now
On Tinker, Inkling 2 Small costs 0.30 dollars per million input tokens and 1.00 dollar per million output tokens, with cached input at 0.03 dollars per million tokens. Artificial Analysis notes the model uses about 150,000 output tokens per index task, so a single answer can take over 20 minutes and cost more per task than the per-token price suggests.
Why this launch matters
A lab founded less than two years ago now has a model at the top of the open-weights rankings. If the weights arrive next week as planned, developers will be able to run, fine-tune, and inspect one of the strongest open models built outside China. The launch also shows how fast the open-weights race is moving, with new models arriving every few weeks and benchmark records falling just as quickly.
Sources
- Thinking Machines Lab - https://thinkingmachines.ai/news/inkling-2-small/ - Artificial Analysis - https://artificialanalysis.ai/articles/thinking-machines-releases-inkling-2-small-highest-scoring-open-weights-model-outside-of-chinia-upon-weights-release-it - MarkTechPost - https://www.marktechpost.com/2026/11/05/thinking-machines-inkling-2-small-release/