Unconfirmed: an X post from Artificial Analysis2 says Thinking Machines' Inkling 2 Small is live on its Tinker platform, and Soumith Chintala1 wrote that open weights follow next week, but Automatica has not confirmed any of it.
Thinking Machines has not been quoted in the evidence we have, and we found no statement from the company outside these posts. Automatica has no earlier reporting on Thinking Machines or Inkling.
Artificial Analysis2 wrote on 5 November that Thinking Machines "has released Inkling 2 Small," available first-party through Tinker and Tinker Playground. It describes the model as proprietary at launch, with an Apache 2.0 weights release planned and the date "unconfirmed." It lists 594B total and 17B active parameters, text and image input, and pricing of $0.30 per million input tokens and $1.00 per million output tokens.
Soumith Chintala1 wrote the same day that "Our v2 small model is here," that it is "Available to train and fine-tune on Tinker right now," and that open weights arrive "next week after external safety testing is fully complete." The post names neither the model nor the company, and the evidence does not establish his affiliation beyond the word "Our." He also claims pre-training took 68 hours on GB300s, post-training ran 17 million rollouts over nine days, and the whole pipeline used NVFP4. We cannot check any of those figures.
The performance numbers come from Artificial Analysis's own evaluation, posted on X. The post says Inkling 2 Small scores 31 on Intelligence Index v5.0.1, up 18 points from Inkling Small, and matches Gemini 3.8 Flash (high) Artificial Analysis2. It reports these gains over Inkling Small:
- AutomationBench-AA: 5% to 46%
- FrontierCode 2.0: 5% to 25%
- HLE-Diamond: 12% to 29%
It also reports a split against Claude Sonnet 5.5 (medium). Inkling 2 Small scores 25% against 11% on FrontierCode 2.0, but 13% against 30% on Terminal-Bench 4.0 Artificial Analysis2.
The cost side matters as much as the scores. The post puts Inkling 2 Small at about 150K output tokens per Intelligence Index task, against about 37K for Inkling Small and about 31K for Sonnet 5.5, and at $1.18 per task against $0.20 for Inkling Small Artificial Analysis2. No methodology, hardware or serving configuration appears in the evidence, and Automatica has not reproduced any of the results.
DesignArena3 wrote that Inkling 2 Small is "4th place overall on DesignArena and is 1st place among open weights models with an Elo of 1376," adding that "Model weights to be released soon." No Elo methodology is given. The ranking also sits oddly with Artificial Analysis's description of a proprietary launch, since no weights are reported as released.
The three accounts also disagree on timing. Chintala says weights come next week, Artificial Analysis says "at a later date," and DesignArena says "soon."
What would confirm the report: a Thinking Machines announcement, a public Tinker listing for the model, or a weights release under Apache 2.0. What would undercut it: a company denial, a different licence, or weights that miss Chintala's stated window. Until then, all three accounts establish only that these claims were made.