Thinking Machines Releases Inkling 2 Small 594B MoE Model
November 7: Vals AI researchers detected profanity in 50 of 2,655 Inkling 2 Small benchmark runs, with 241 instances from two spiraling runs.
Thinking Machines released Inkling 2 Small, a 594B MoE with 17B active parameters ranking 31 on Artificial Analysis index; open weights next week after safety testing.
Thinking Machines released Inkling 2 Small, a mixture-of-experts model with 594 billion total parameters and 17 billion active ones. Trained from scratch in under 3 days on 8.4T-12T tokens in NVFP4, it supports 1M context and is available for fine-tuning. The model ranks 31 on the Artificial Analysis Intelligence Index, on par with Google's Gemini 3.8 Flash, with gains in agentic tasks. Open weights are planned under Apache 2.0 after external safety testing. Researchers found profanity in 50 of 2,655 benchmark runs, including 316 instances from two spiraling runs.
- 594B total parameters, 17B active
- Trained in 68 hours on 12T tokens
- 1M context window
- Ranks 31 on Artificial Analysis Intelligence Index
- Open Apache 2.0 weights next week after safety testing
- Profanity in 50 of 2,655 runs
Timeline
dated events in the reporting · oldest first- Nov 6, 2026Thinking Machines releases Inkling 2 Small MoE · MarkTechPost, Latent Space, The Decoder
- Nov 7, 2026Researchers find profanity in 2,655 benchmark runs · The Decoder
What each outlet adds
beyond the shared facts- Latent SpaceTraining details: 68 hours on 12T tokens, 1M context, fine-tuning support, Apache 2.0 weights next week
- The DecoderTop spot among Western open-weight models, 31 on index, $1.18 per task cost
- The DecoderProfanity in 50 of 2,655 runs, 316 instances from two spiraling runs
- Products
- Inkling 2 Small
- Organizations
- Artificial Analysis
- AI models
- Inkling 2 Small
Coverage
oldest first · one row per article · UTC- · 3d ago
MarkTechPostFirst report
Thinking Machines Releases Inkling 2 Small model
The 594 billion parameter Mixture-of-Experts model has 17 billion active parameters.
- · 3d ago
Latent Space
Thinky Machines releases Inkling 2 Small, 594B MoE beating Muse Spark
Apache 2.0 weights ship next week after safety testing; ranks 31 on Artificial Analysis Index.
- · 3d ago
The Decoder
Thinking Machines Inkling 2 Small claims top spot among Western open-weight models
Inkling 2 Small scores 31 on Artificial Analysis Intelligence Index, 18 points above predecessor, at higher cost.
- · 1d ago
Inkling 2 Small produces 316 profanity instances in benchmark runs
Researchers at Vals AI found profanity in 50 of 2,655 model runs across 13 benchmarks, with 241 instances from two spiraling runs.