A technical analysis by an unnamed author argues that nondeterminism in LLM inference is not primarily caused by GPU concurrency and floating-point non-associativity, as commonly hypothesized, but by …
Tinker's LoRA fine-tuning matches full fine-tuning performance on small datasets and reinforcement learning, but underperforms on larger supervised learning datasets, according to a technical primer f…
Tinker, an AI model fine-tuning platform, has added GLM 5.3 (256K) to its training lineup, alongside other models such as Inkling, Nemotron-3.5-Lightning-30B-A3B, and Qwen3.8-27B, with prices per mill…
A fine-tuned model called ReViSQL-K2.6, developed by Tinker using reinforcement learning with verifiable rewards, exceeds the human benchmark of 92.96% on the BIRD text-to-SQL task, reaching 92.96% wh…
Thinking Machines released Inkling and Inkling-Small, two open-weight language models, and outlined a safety framework for releasing open weights, emphasizing robust safety testing and ecosystem readi…
Thinking Machines Lab, Inc. released Inkling, a general-purpose multimodal model with 975 billion total parameters and 41 billion active parameters, on July 15, 2026 under an Apache 2.0 license. The m…
Murati's Thinking Machines released Inkling, an open-weights 975B parameter mixture-of-experts LLM with 41B active parameters, supporting text, image, and audio input with a 1M-token context window. T…
Inkling, a new open-weights AI model with 975B total parameters and 41B active parameters, has been released by an unnamed organization to advance its mission of building AI that extends human will an…
Thinking Machines, an AI company, argues that artificial intelligence should be designed to extend human will and judgment rather than replace it, emphasizing the need for distributed, continuously le…
A proprietary model trained on high-quality human annotations outperforms all frontier models on financial information filtering tasks, achieving over 80% accuracy at a fraction of the cost, while Gem…