The architecture features 594 billion total parameters with 17 billion active, offering support for variable computational effort alongside native image inputs. Tailored for agentic coding and autoresearch, it was refined on hundreds of thousands of programming challenges paired with an agentic grader to master GPU kernel optimization and autonomous machine learning. Benchmark evaluations show it secured 72.1% on DeepSWE v1.1 and 86.3% on SWEBench Pro v2 (Hard). Following safety evaluations, downloadable checkpoints will be made publicly available next week under an Apache 2.0 license.
Key points
The system is currently accessible for sampling and fine-tuning with closed weights on Ti…: The system is currently accessible for sampling and fine-tuning with closed weights on Tinker Playground.
Early developmental checkpoints of the model were evaluated on OpenCode using the identif…: Early developmental checkpoints of the model were evaluated on OpenCode using the identifier Fledge-Alpha.
Full model weights will be released publicly next week under an Apache 2.0 license after…: Full model weights will be released publicly next week under an Apache 2.0 license after safety evaluations conclude.
Sources checked by our analysis pipeline: 1
Today we are introducing our second model generation with Inkling 2 Small. It is a Mixture-of-Experts model with 594B total parameters, 17B active, and supports image inputs and variable effort.
Inkling 2 Small far surpasses our previous models’ scores on the Artificial Analysis Intelligence Index with a score of 31, reflecting gains in performance on agentic coding, vision, knowledge work, and reasoning. The increase in capabilities comes from improvements made at every step of our training process, including more efficient pre-training and a large, unified RL run.
Inkling 2 Small will initially be available through Tinker, with closed weights, following the approach we outlined in a safe path to open weights. We intend to make the full weights available under an Apache 2.0 license next week after the completion of external safety testing and red-teaming.
You can try Inkling 2 Small on Tinker Playground today. The model is also available for fine-tuning and sampling on Tinker; checkpoints will be downloadable with the release of Inkling 2 Small’s weights.
Capabilities
Inkling 2 Small is more capable across the board than our…
The artificial intelligence market is recalibrating expectations as Anthropic faces pressure to slash its initial public offering valuation in half to $1 trillion. This contraction stems from plateauing quarterly revenues and persistent training expenses that outpace monetization. The adjustment threatens to trigger a downstream financing crunch for rivals like OpenAI, whose own high-flying private valuations relied on Anthropic maintaining a far richer multiple.
Joint guidance from three national agencies follows the rollout of autonomous tools like Meta’s Muse, which can independently browse the web, handle emails, and execute bookings. Officials warn that granting unchecked permissions invites severe security breaches, particularly because users bear full legal liability for any tasks their assistants execute. Consumers must establish rigorous guardrails to keep these evolving digital systems separated from sensitive personal records.
A reported data exposure behavior has emerged regarding the Claude Code platform and user privacy. External reference material documenting the transmission mechanism is currently available for review through an online repository. The revelation originated from a community report raising visibility around automated software behaviors.