Thinking Machines has released Inkling 2 Small, Artificial Analysis said. For now it runs only on the company's Tinker platform and Tinker Playground. Of its 594 billion parameters (the numbers a model learns), only 17 billion are active at a time, so a small slice works on each step. Artificial Analysis rates it 18 points above Inkling Small on its Intelligence Index, level with Google DeepMind's Gemini 3.8 Flash (high).
The weights are not out yet. Artificial Analysis said an Apache 2.0 release is planned, with no confirmed date. Its hallucination rate fell from 61% to 27%, but mostly because it answers less. It attempted 53% of questions, down from 77%. Accuracy rose only from 32% to 34%. It also writes a lot while reasoning: about 150,000 output tokens per task, four times Inkling Small.
Per-task cost is the fair yardstick for a model this wordy. Artificial Analysis put Inkling 2 Small at $1.18 per task, well under Gemini 3.8 Flash (high) at $7.77 and Claude Sonnet 5.5 (medium) at $10.02. Coding is mixed: 25% versus 11% for Claude Sonnet 5.5 (medium) on FrontierCode 2.0, but 13% versus 30% on Terminal-Bench 4.0. Buyers should ask: does our work resemble the test it wins? Can we accept more skipped questions? Can we plan around hosted-only access until weights arrive?