Thinking Machines' Inkling 2 Small scores 31 on AA Intelligence Index, eyeing top open-weights spot outside China
ArtificialAnlys · x · 2026-11-06
Artificial Analysis reviewed Thinking Machines' newly released Inkling 2 Small, which scores 31 on the Intelligence Index v5.0.1 (up 18 from Inkling Small), matching Gemini 3.8 Flash (high). If weights are released as planned under Apache 2.0, it would be the highest-scoring open-weights model developed outside China.
Key facts
- 594B total / 17B active parameters; priced at $0.30/$1.00 per 1M input/output tokens ($0.03 cached); launching proprietary via Tinker and Tinker Playground, weights to follow.
- Big agentic gains: AutomationBench-AA 5%→46%, FrontierCode 2.0 5%→25%, HLE-Diamond 12%→29%.
- Complementary coding profile to Claude Sonnet 5.5 (medium): 25% vs 11% on FrontierCode 2.0, but 13% vs 30% on Terminal-Bench 4.0.
- Creative agentic tasks: AA-3D 1021 Elo (+438, above Gemini 3.8 Flash's 966); Code2Video-AA 1020 Elo (+553, near Qwen3.8 Max).
- Long-horizon business workflows: AA-Briefcase 1527 Elo (+615), 50% rubric pass rate, beating Sonnet 5.5 (1439) and Gemini 3.8 Flash (1203).
- Hallucination rate fell 61%→27% mainly via more selective answering (attempt rate 77%→53%); accuracy only 32%→34%.
- 150K output tokens per task (4x its predecessor); $1.18 per task, 6x costlier than Inkling Small but far cheaper than Gemini 3.8 Flash ($7.77) or Sonnet 5.5 ($10.02).
More from Models
- Thinky Machines launches Inkling 2 Small, scoring 31 on AA Intelligence Index — LiTianleli · 2026-11-06
- Pre-training done in 68 hours with 4-bit used all the way from pre- to post-training — teortaxesTex · 2026-11-06
- METR: No clear bottlenecks seen, capability progress could accelerate further in 12 months — paulnovosad · 2026-11-06
- 18 months from o3 to Claude Mythos: public models still close to last year's frontier — louisvarge · 2026-11-06
- Tencent Hunyuan Hy4 takes #1 Chinese model spot on Agent Arena at $0.57 per task — arena · 2026-11-06
- Agent Arena leaderboard: Tencent's Hy4 leads Chinese models at $0.57/task — arena · 2026-11-06