Release2 sources
[AINews] Inkling 2 Small: 17A-594B trained in 68 hours on 12T tokens, beating Muse Spark
Latent Space · By swyx .ai ·
Build AI systems with context graphs and agentic memory. Join Andrew Ng and AI engineers at NODES 2026 for 100+ live technical sessions . Free, online, 24 hours. Save your spot ! There was a time in the early days of Thinking Machines, post 2025 start and cofounder departures…
The Brief
- Thinking Machines has released Inkling 2 Small, per a Hacker News front-page link to its announcement.
- Latent Space reports the model is described as 17A-594B, trained in 68 hours on 12T tokens.
- Latent Space reports that Inkling 2 Small beats Muse Spark; the sources give no benchmark details.
Why it matters
Developers comparing open or hosted models may want to check the primary announcement for the architecture, training details and benchmark results, since the available sources contain few specifics.
Who’s affected
- Developers evaluating large language models
- #llm
- #model-release
- #ai-models
- #thinking-machines
- #benchmarks
Timeline
Inkling 2 Small announced by Thinking Machines
Hacker News: Front Page
Inkling 2 Small
Latent Space
[AINews] Inkling 2 Small: 17A-594B trained in 68 hours on 12T tokens, beating Muse Spark
Full coverage · 2 sources
- 2h ago ↗
Latent Space
[AINews] Inkling 2 Small: 17A-594B trained in 68 hours on 12T tokens, beating Muse Spark
- 12h ago ↗
Hacker News: Front Page
Inkling 2 Small
