Thinking Machines Lab Releases Inkling 2 Small, a 594B Model With Open Weights Promised

Thinking Machines Lab on Wednesday released Inkling 2 Small, the second-generation model from the startup founded by former OpenAI chief technology officer Mira Murati — a model that Artificial Analysis says would become the strongest open-weight model developed outside China once its weights are released.

The model is a mixture-of-experts system with 594 billion total parameters and 17 billion active. It accepts image inputs, supports variable reasoning effort, and was trained with reinforcement learning on hundreds of thousands of long-horizon coding tasks across multiple agent harnesses. It runs on a 256K-token context window, with compaction for longer trajectories.

Benchmark performance

Artificial Analysis scores Inkling 2 Small at 31 on its Intelligence Index v5.0.1 — 18 points above the original Inkling Small and level with Gemini 3.8 Flash at high effort. The lab says the model jumps 41% in automation capabilities and 20% in coding abilities over its predecessor.

On coding benchmarks, it scores 25.4% on FrontierCode v2.0, 72.1% on DeepSWE v1.1 and 86.3% on SWEBench Pro v2 (Hard). In external GPU kernel optimisation work, it solved 54.25 of 69 tasks on PMPP-Hard, behind GPT-6 Astra (63) and Kimi K3.1 Pro (60.5).

The model also shows strength in vision. It uses a 292-million-parameter vision encoder and leads its comparison table on benchmarks including MMMU Pro, PerceptionBench and ScreenSpot Pro.

Agentic training

Thinking Machines says the gains come from improvements at every stage of training, including more efficient pre-training and a large, unified reinforcement-learning run. Notably, many tasks used an agentic RL grader that evaluated how the model approached a problem, not just whether its code passed unit tests.

The lab demonstrated the model’s autonomous research ability with a time-series forecasting task: given a Qwen3.5-4B base model and $426 of compute budget, Inkling 2 Small wrote its own evaluator, curated data, ran supervised fine-tuning and reinforcement learning, and more than doubled the model’s forecasting score in 4.5 hours.

Open weights, with caveats

Inkling 2 Small is initially available through Tinker with closed weights, following what the lab calls a safe path to open weights. The company says it intends to release the full weights under an Apache 2.0 licence next week, after external safety testing and red-teaming are complete.

API pricing is $0.30 per million input tokens and $1.00 per million output tokens, though the model consumes roughly 150,000 output tokens per index task — about four times its predecessor.

The lab is also in the news for its financing. Business Insider reported that Thinking Machines is raising $5 billion at a $40 billion pre-money valuation, led by Accel, which would more than triple the valuation it reached after a $2 billion seed round last year. The company declined to comment.

Inkling 2 Small has limitations. It does not support audio, is less optimised for chat, and the lab acknowledges weaker organisation of content and occasional overreaction to safety-related queries.

Key points

• 594B total parameters, 17B active, image input, variable reasoning effort.
• Artificial Analysis Intelligence Index score of 31, up 18 points on its predecessor.
• Full Apache 2.0 weights promised next week after external safety testing.
• Reportedly raising $5 billion at a $40 billion pre-money valuation.