Clarity for AI buildersSponsored by AI Routing
availability

Thinking Machines launches Inkling 2 Small with higher benchmark scores; open weights still pending

Thinking Machines launches Inkling 2 Small with higher benchmark scores; open weights still pending
Thinking Machines has launched Inkling 2 Small as a proprietary reasoning model through its Tinker platform and Tinker Playground, according to Artificial Analysis. The model scores 31 on the Artificial Analysis Intelligence Index, an 18-point improvement over Inkling Small. An Apache 2.0 weights release is planned, but its date is unconfirmed. Artificial Analysis says the planned release would make Inkling 2 Small the highest-scoring open-weights model developed outside China, based on its current rankings. That distinction remains conditional: the weights are not available at launch, and the comparison reflects the rankings reported in the source. On the Artificial Analysis Intelligence Index v5.0.1, Inkling 2 Small matches Google DeepMind’s Gemini 3.8 Flash (high) and surpasses Xiaomi’s MiMo-V2.6-Pro. The reported gains over Inkling Small are particularly large on agentic evaluations: AutomationBench-AA rises from 5% to 46%, FrontierCode 2.0 from 5% to 25%, and HLE-Diamond from 12% to 29%. The coding results are mixed against Claude Sonnet 5.5 (medium). Inkling 2 Small scores 25% on FrontierCode 2.0, compared with 11% for Sonnet, but reaches 13% on Terminal-Bench 4.0 against Sonnet’s 30%. On the research-level physics evaluation CritPt v2, it improves from Inkling Small’s 34% to 56%, while remaining below Gemini 3.8 Flash (high) at 65%. Artificial Analysis also reports stronger performance on long-horizon business workflows. On AA-Briefcase, which tests tasks involving spreadsheets, presentations and memos, Inkling 2 Small scores 1527 Elo and passes 50% of rubric checks. Inkling Small scores 912 Elo and passes 20%. The reported comparison scores are 1439 Elo for Claude Sonnet 5.5 (medium) and 1203 Elo for Gemini 3.8 Flash (high). In creative agentic tasks, Inkling 2 Small scores 1021 Elo on AA-3D, which evaluates the creation of interactive 3D experiences, and 1020 Elo on Code2Video-AA, which evaluates video creation using code. Artificial Analysis reports that it exceeds Gemini 3.8 Flash on both evaluations but trails Claude Sonnet 5.5 on both. The reduction in hallucinations comes with a qualification. On AA-Omniscience v1.1, the reported hallucination rate falls from Inkling Small’s 61% to 27%, but the attempt rate also drops from 77% to 53%. Accuracy increases only slightly, from 32% to 34%. Artificial Analysis attributes the lower hallucination rate mainly to more selective answering rather than a large improvement in accuracy. Higher benchmark performance also requires substantially more generated text. Inkling 2 Small averages approximately 150,000 output tokens per Intelligence Index task, compared with approximately 37,000 for Inkling Small, 99,000 for Gemini 3.8 Flash (high), and 31,000 for Claude Sonnet 5.5 (medium). Artificial Analysis puts Inkling 2 Small’s cost at $1.18 per Intelligence Index task—almost six times Inkling Small’s $0.20, though below the reported $7.77 for Gemini 3.8 Flash (high) and $10.02 for Claude Sonnet 5.5 (medium). These figures describe costs on that evaluation, not a general estimate for every workload. The source lists Inkling 2 Small as having 594 billion total parameters and 17 billion active parameters, with text and image input and text output. Token pricing is $0.30 per million input tokens, $1.00 per million output tokens, and $0.03 per million cached input tokens. For now, access is through Tinker and Tinker Playground; the planned weights release remains pending.