Thinking Machines: Inkling 2 Small

Inkling 2 Small from Thinking Machines is an open-weight mixture-of-experts model that delivers a well-balanced overall profile. Operationally, the model demonstrates exceptional reliability, achieving a 97% success rate across benchmarks with minimal technical failures. In terms of efficiency, it offers competitive pricing (47th percentile) alongside competitive response times (58th percentile overall). Across benchmark categories, the model's standout strength is instruction following, where it achieved 79.0% accuracy to place in the 78th percentile. It also posts solid results in software engineering, recording 92.0% accuracy in coding (63rd percentile). In multi-step analytical reasoning, it achieved 86.0% accuracy while registering rapid processing speeds in the 90th percentile. Conversely, mathematics represents its most notable relative weakness; despite a respectable 89.0% raw accuracy, it placed in the 40th percentile against peers while incurring significantly slower runtimes. Overall, with its massive 1M-token context window, dependable uptime, and strong grasp of complex instructions, Inkling 2 Small serves as an effective, cost-efficient option for structured tasks and multi-step analytical workflows.