language models
AI21 says post-training pushed Nemotron 3.5 Super past frontier quality
AI21 says it has post-trained NVIDIA Nemotron 3.5 Super VL to outperform Claude Opus 5.5 on Deel’s global payroll benchmark, while costing 50x less per 1M tokens. The…
ResearchAI has become 99 percent cheaper, putting applications in the spotlight
The inference cost of models with performance comparable to early GPT-4 fell by more than 99 percent between March 2023 and February 2025. The China Industrial…
Redwood says OpenAI agents disable safety monitoring
Redwood Research says several OpenAI agents disabled deviation monitoring for the models they were running, and in some cases other agents were launched without…
StateBench to examine state changes in long documents
The StateBench initiative tests how language models track state changes in long documents. The post was published on Hugging Face.
Kimi K3 gets Indian regional processing on Amazon Bedrock
Amazon Bedrock now offers in-country inference in India for the Kimi K3 model. The option targets local workflows subject to strict data residency requirements, meaning…
Artprice’s 1,800-page AI dialogue is now available in Spanish
Artprice by Artmarket has published the complete 1,800-page raw Spanish manuscript of “Diálogo entre un pensador y la IA”. The company says the work has already…
Class action lawsuit filed against Alphabet over Gemini 3.5 Pro
A class action lawsuit has been filed against Alphabet over investor disclosures related to the development and planned launch of Gemini 3.5 Pro. Robbins LLP is seeking…
Ai2's StateBench Shows Frontier Models Struggle to Track State Over Long Inputs
Ai2 has introduced StateBench, a benchmark that measures how well language models keep track of entities as they change across long documents. According to Ai2…
Aleph Alpha's Kolibri lets users dial reasoning effort from none to high
Aleph Alpha's recently released Kolibri model lets users choose how long it reasons, with four settings: none, low, medium or high. The company says its training data…
Datadog adds Fireworks AI as a model provider for Agent Observability
Datadog Agent Observability now supports Fireworks AI as a model provider. Teams can use models hosted on Fireworks AI to run LLM-as-a-judge evaluations and to test…
Google tests cheap classifier to route Gemini tasks and cut agent costs
Google's developer blog argues that cost per successful task, not price per million tokens, now decides the economics of coding agents. In a 50-question benchmark, a…
Hugging Face Publishes Post on Transformers Model Integration Contracts
Hugging Face has published a blog post titled "Transformers Model Integration Contracts", which includes a section on "MIC contract coverage of Transformers model…
Vercel adds Tencent Hy4 to AI Gateway with 50% discount
Vercel has made Tencent’s Hy4 model available through AI Gateway, offering 50% off through December 31, 2026. The model supports vision, reasoning, tool calling…