Thinking Machines Launches Inkling Small: Open-Weights Model Matches Flagship at 1/3 Size
ArtificialAnlys · x · 2026-07-31
Thinking Machines, the AI lab founded by former OpenAI CTO Mira Murati, has released Inkling Small, its second model. It is an open-weights reasoning model (Apache 2.0) with 276B total parameters and 12B active parameters (MoE), supporting text, image, and speech inputs with a 256K context window.
Key Benchmark Results:
- Intelligence Index: Scores 40 on the Artificial Analysis Intelligence Index, within a point of its flagship sibling Inkling (41), despite having less than one-third of the parameters. No open-weights model at its size or smaller scores higher.
- Strengths: Meets or exceeds the flagship on several coding and frontier reasoning evals, including Humanity's Last Exam (32% vs. 30%), GPQA Diamond (89% vs. 87%), and SciCode (49% vs. 46%).
- Weaknesses: Trails the flagship on agentic tasks and factual knowledge. Scores -9 on the AA-Omniscience Index (vs. 2 for the flagship), driven by lower accuracy (31% vs. 40%), though its hallucination rate is slightly lower.
- Token Efficiency: Averaged 24K output tokens per task, slightly fewer than the flagship, and significantly less than peers at its intelligence level like DeepSeek V4 Flash (45K) and GPT-5.4 mini (78K).
Related event: Thinking Machines Unveils Inkling-Small Model(22 posts)→
More from Models
- Dev Questions if Opus 5 Cheated on Benchmarks, Calls it Unusable for ML — ostrisai · 2026-07-31
- Transluce Releases WeirdChat: A Catalog of 175K Strange LLM Behaviors — ChowdhuryNeil · 2026-07-31
- Chart Shows LLMs Offer Incredible Intelligence Per Dollar — downingARK · 2026-07-31
- GPT-5.6 Hits 13% in AI Space Race Test, Beating Open-Source Kimi K3 — scaling01 · 2026-07-31
- OpenAI's GPT-5.6 Self-Optimizes: Slashes Serving Costs by 20% — tszzl · 2026-07-31
- Scholars Debate Scaling Laws: Are Models Less General Despite Growing Stronger? — davidmanheim · 2026-07-31