Inkling 2 Small Tops US Open Models in Agent Arena
The cost gap is the real story: at US$0.26 per agentic task, Inkling 2 Small undercuts Kimi K3.1 Pro by nearly 5x, which changes the economics of running long multi-step agent workflows. Teams building coding or agent pipelines who previously defaulted to Chinese open-weights models now have a US-hosted option to benchmark. Developers can test it directly in Agent Arena and Code Arena before committing to a migration.
Thinkymachines released Inkling 2 Small with open weights, and the model immediately took the #1 spot among US open models in Agent Arena, ranking #2 among all open models worldwide with a live +2.9% net improvement score across 11,500 real-world agentic sessions (#22 overall). It is also the only US model inside the Agent Arena open top 10.
The model shows +10.9% Steerability, the highest of any open model, and a median cost of US$0.26 per task, roughly 4.7x cheaper than the #1 open model, Kimi K3.1 Pro from Moonshot (Kimi), which costs US$1.22 per task. In the Code Arena: WebDev track, Inkling 2 Small scores 1,647 points, making it the #1 US open model there as well, #4 among open models globally and #23 overall. That is a +239-point jump over its predecessor Inkling-Small, which scored 1,408 points and ranked #34 among open models.
Agent Arena evaluates models using real usage traces rather than human preference votes, aiming for objective signals on agentic performance from a global user community. The dual #1 rankings in both agentic and web-dev coding tasks signal that US open-weights models are narrowing the gap with Chinese leaders like Kimi on cost and capability.
With its open weights release today, Inkling 2 Small by @thinkymachines is now also the #1 US open model in Agent Arena, and #2 among all open models. It has a live +2.9% net improvement score across 11.5k real-world agentic sessions, ranking #22 overall. Some more stats: - +10.9% Steerability: the highest of any open model - $0.26 median cost per task: about 4.7x less than the #1 open model, Kimi K3.1 Pro ($1.22) - The only US model in the Agent Arena open top 10 In the Agent Arena, we use real usage traces for objective signals to move beyond human preference. Everything in the Arena is evaluated based on real-world usage with a global community of users. Arena.ai @arena Exciting news: Inkling 2 Small by @thinkymachines is now officially open source, and currently in the Code Arena: WebDev with 1647 pts, making it the #1 US open model! Inkling 2 Small also ranks #4 among open models globally and #23 overall. It's a +239 pt improvement over its previous variant, Inkling-Small (1408 pts, #34 among open models). Congrats to @thinkymachines for their contribution to open progress! 🔗 View Quoted Tweet
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Inkling 2 Small, Thinkymachines, Kimi K3.1 Pro, Agent Arena.
Source attribution: curated from lmarena.ai (@lmarena_ai). Cross-source corroboration: 4 independent sources have covered this development (including Thinking Machines Lab, vLLM, Together AI), confirming industry-wide signal strength.
