Thinking Machines Releases Inkling 2 Small With a Score of 31 on the Artificial Analysis Intelligence Index

Thinking Machines Releases Inkling 2 Small With a Score of 31 on the Artificial Analysis Intelligence Index

Thinking Machines has released Inkling 2 Small, a proprietary reasoning model that scores 31 on the Artificial Analysis Intelligence Index v5.0.1. The company plans to release its weights under the Apache 2.0 license at a later date, although no release date has been confirmed. Based on current rankings, that release would make Inkling 2 Small the highest-scoring open weights model developed outside China.

At launch, Inkling 2 Small is available through Thinking Machines' Tinker platform and Tinker Playground. Its Intelligence Index score is 18 points higher than Inkling Small's, while the model generates approximately four times as many output tokens per task. It matches Google DeepMind's Gemini 3.8 Flash (high) and scores above Xiaomi's MiMo-V2.6-Pro.

Key results

• Significant gains over Inkling Small: AutomationBench-AA rises from 5% to 46%, FrontierCode 2.0 increases from 5% to 25%, and HLE-Diamond improves from 12% to 29%.
• Different coding results from Claude Sonnet 5.5 (medium): Inkling 2 Small scores 25% on FrontierCode 2.0, compared with 11% for Claude Sonnet 5.5 (medium). On Terminal-Bench 4.0, however, it scores 13%, compared with 30% for Claude Sonnet 5.5 (medium).
• Strong creative agent performance: On AA-3D, Inkling 2 Small scores above Claude Haiku 5.5 and Grok 4.7. It also scores above Gemini 3.8 Flash on AA-3D and Code2Video-AA, while remaining behind Claude Sonnet 5.5 on both evaluations.
• Improved physics performance: Inkling 2 Small scores 56% on CritPt v2, up from 34% for Inkling Small. Gemini 3.8 Flash (high) scores 65%.
• Lower hallucination rate: On AA-Omniscience v1.1, the hallucination rate falls from 61% for Inkling Small to 27%. Accuracy rises from 32% to 34%, while the attempt rate drops from 77% to 53%.
• Higher output-token usage: Inkling 2 Small averages approximately 150K output tokens per Intelligence Index task, compared with roughly 37K for Inkling Small, 31K for Claude Sonnet 5.5 (medium), and 99K for Gemini 3.8 Flash (high).
• Higher task cost: Each Intelligence Index task costs $1.18 for Inkling 2 Small, compared with $0.20 for Inkling Small, $7.77 for Gemini 3.8 Flash (high), and $10.02 for Claude Sonnet 5.5 (medium).

Model details

• Type: Proprietary reasoning model at launch, with an Apache 2.0 weights release planned
• Modalities: Text and image input, text output
• Size: 594B total parameters, with 17B active parameters
• Pricing: $0.30 per 1M input tokens, $1.00 per 1M output tokens, and $0.03 per 1M cached input tokens
• Availability: Tinker and Tinker Playground

Business workflow evaluation

On AA-Briefcase, Inkling 2 Small scores 1527 Elo, 615 points higher than Inkling Small's 912 Elo. The evaluation measures long-horizon business workflows involving spreadsheets, presentations, and memos.

The model also scores above Claude Sonnet 5.5 (medium), at 1439 Elo, and Gemini 3.8 Flash (high), at 1203 Elo. Inkling 2 Small passes 50% of rubric checks, compared with 20% for Inkling Small. Its analytical quality score is 1634 Elo, and its presentation score is 1564 Elo.

Creative and code-generated video tasks

Inkling 2 Small makes substantial gains over Inkling (xhigh) on AA-3D and Code2Video-AA. On AA-3D, which evaluates the ability to create interactive 3D experiences, Inkling 2 Small scores 1021 Elo. This compares with 966 Elo for Google DeepMind's Gemini 3.8 Flash and 1033 Elo for Meta's Muse Spark 1.3 (max).

On Code2Video-AA, which evaluates video creation using code, Inkling 2 Small scores 1020 Elo. Gemini 3.8 Flash scores 899 Elo, while Alibaba's Qwen3.8 Max (0902) (xhigh) scores 1021 Elo.

Compared with Inkling (xhigh), Inkling 2 Small gains 438 Elo on AA-3D, increasing from 583 to 1021, and 553 Elo on Code2Video-AA, increasing from 467 to 1020.

Hallucination and answer selection

Inkling 2 Small's improved AA-Omniscience v1.1 result is associated primarily with answering fewer questions. Its net score rises to 16, compared with -9 for Inkling Small, but remains below Gemini 3.8 Flash (high), at 23, and Claude Sonnet 5.5 (medium), at 19.

The hallucination rate declines from 61% to 27%, while the attempt rate falls from 77% to 53%. Accuracy improves only slightly, from 32% to 34%.

Output-token usage and pricing

Inkling 2 Small averages approximately 150K output tokens for each Intelligence Index task, nearly four times Inkling Small's approximately 37K. It also uses more output tokens than Gemini 3.8 Flash (high), at approximately 99K, and Claude Sonnet 5.5 (medium), at approximately 31K.

The model costs $1.18 per Intelligence Index task, nearly six times the $0.20 cost for Inkling Small. Its token pricing is:

• $0.30 per 1M input tokens
• $1.00 per 1M output tokens
• $0.03 per 1M cached input tokens

The complete results cover the evaluations included in the Artificial Analysis Intelligence Index v5.0.1.