Source categories
New models, APIs and product releases from OpenAI, Google DeepMind and Anthropic, plus Model Context Protocol specification updates.
15 related activitiesLast 30 days 12 itemsFeatured among them 1 items
Terminal X matches the judgment accuracy of seasoned investors at 88.6% with GPT-6.1. Its Eval Agent rates outputs against a 54-criteria rubric, and with GPT-6.1 Sol it can identify 25% more anomalies that deviate from fund mandates.
Read originalOpenAI is improving ChatGPT’s code generation performance for low-resource programming languages, adding three evaluation dimensions: software engineering, SQLite, and COBOL migration. These capability improvements are centered on code generation scenarios for low-resource programming languages.
Read originalItaú is implementing its AI First strategy in customer and employee services through the conversational interface ia.i and the IARA governance platform. This application scenario operates in one of Brazil's most heavily regulated industries.
Read originalBenefit Street Partners speeds up question answering with OpenAI's Codex tool. The firm's Anant Kumar uses Codex to conduct code reviews, improve follow-up efficiency, and build tools to compare loan valuations and analyze credit agreements.
Read originalSullivan & Cromwell has partnered with OpenAI to develop Agreement Analyzer, an AI tool designed to identify issues in draft agreements. The tool integrates legal expertise into the agreement review process.
Read originalAmp leverages OpenAI's GPT-6 Astra, Ultrafast, and GPT-Live-1 models, along with Sign in with ChatGPT, to expand AI agent use cases beyond coding into other domains. This practice aims to explore the capability boundaries of AI agents.
Read originalAnthropic opened an office in Singapore, its fifth in the Asia-Pacific region, and signed a memorandum of understanding with Singapore's Ministry of Digital Development and Information. The two sides will cooperate in areas such as AI safety, model evaluation, and responsible AI implementation. Singapore is one of Anthropic's most active global markets, and Claude has been deployed in multiple local sectors including government, finance, and technology.
Read originalGoogle DeepMind is advancing the widespread deployment of Gemini 4 Argon while expanding the coverage of trusted cyber defenders in the Fairwind Program. The model solved all 12 scenarios in end-to-end simulated cyberattack tests, a significant improvement from 5 in the previous generation. Its refusal rate for violating requests nearly doubled compared to the previous generation, and violations accounted for only 2% of attempts in coding benchmark tests.
Read originalGoogle DeepMind officially released the frontier large model Gemini 4 Argon on October 13, 2026, available globally to developers, enterprises, researchers, and consumers. The model ranks first on the Vals Index benchmark, with input token pricing at $2 per million and output tokens at $10 per million, with a 95% discount for cached input tokens. It has already been deployed in multiple practical applications across engineering optimization, quantum chemistry, cybersecurity, and other fields.
Read originalAnthropic has released an update to its usage policy for 2026, which will take effect on November 12. This update mostly clarifies existing rules, adds integrated specialized provisions for deceptive activities, refines restrictions related to elections, weapons, and surveillance, specifies requirements for high-risk use cases, and adds prohibitions against malicious behavior targeting models. The full policy is available on the corresponding page.
Read originalGoogle DeepMind released EmbeddingGemma 2 on October 6, 2026, based on the Gemma 4 architecture, open-sourced under the Apache 2.0 license, with a total parameter count of 740 million, capable of mapping text, code, images, audio, and video into a unified embedding space.
Key Takeaways:Learn about EmbeddingGemma 2's modular parameter configuration, vector truncation rules, and on-device memory usage for building local multimodal retrieval and RAG pipelines.
Google DeepMind announced the launch of the frontier large model Gemini 4 Argon on September 30, 2026. The model is designed for complex long-cycle workflows such as software engineering, legal and financial enterprise knowledge work, and cybersecurity defense, and is currently being phased in to trusted cyber defenders through the Fairwind Program.
Read originalModel Context Protocol has released the 2026-07-28 Release Candidate (RC) version. The current specification is provided as a draft, and detailed changes can be found in the corresponding draft changelog. This specification is not yet finalized and may still be adjusted between the RC and official release. Each SDK will adapt to this version at its own pace, and older versions of the specification may continue to be used for an unspecified period.
Read originalThe Model Context Protocol 2026-07-28 revision has been officially released as a stable version. The full specification is available on the official Model Context Protocol website, and detailed changes can be found in the corresponding version's changelog.
Read originalThe Model Context Protocol 2025-11-25 revision has been officially released. The specification is available on the official Model Context Protocol website, and detailed changes can be found in the corresponding version's changelog.
Read original
