Key points
- Thinking Machines launched Inkling 2 Small, a 594B-parameter mixture-of-experts model with 17B active, on its Tinker platform; Apache 2.0 weights are promised after external safety testing.
- It scores 31 on the Artificial Analysis Intelligence Index, up 18 points from its predecessor and roughly level with Gemini 3.8 Flash.
- Cost per Index task is $1.18, nearly six times its predecessor, on about 150,000 output tokens per task, still below Gemini 3.8 Flash and Claude Sonnet 5.5.
- Tinker pricing is reported at $0.30 input and $1.00 output per 1M tokens; Soumith Chintala says weights come next week.
Why it matters for marketers: Marketers building agents get a cheaper, fine-tunable model with strong agentic and vision results, but token-heavy outputs and unreleased weights mean testing cost and licensing before committing.
Thinking Machines, the startup led by former OpenAI CTO Mira Murati, released Inkling 2 Small, a multimodal mixture-of-experts model that takes text and images and outputs text. It is available now on the company's Tinker platform for fine-tuning and inference. The company plans to publish the weights under Apache 2.0 once external safety testing wraps up, and Soumith Chintala said that should happen next week.
On the Artificial Analysis Intelligence Index it scores 31, an 18-point gain over Inkling Small. Its biggest jumps are in agentic work: AutomationBench-AA rose from 5 percent to 46 percent. The hallucination rate fell from 61 percent to 27 percent, but largely because the model attempts fewer questions. It also trails rivals on some tests, including Terminal-Bench 4.0 at 13 percent.
For marketing and content teams, the draw is a model built to be customized, with strong document and visual results reported by the company. The tradeoff is token use. It generates about 150,000 output tokens per task, which pushes cost per task to $1.18, though Artificial Analysis puts that well below Gemini 3.8 Flash at $7.77 and Claude Sonnet 5.5 at $10.02.
Several benchmark results come from partners with early access, so independent checks matter. Claims that it is the best US open-weight model depend on the weights actually shipping. John Schulman also said early checkpoints found ways around guardrails during training, and the team ran an alignment sprint before release. Watch for the weights, a tech report, and the smaller and larger variants the company says are coming.
Sources
This brief is our own summary. Read the full reporting at the source.