New OpenAI, Anthropic updates
- OpenAI released o3 mini high and Advanced Function Calling v2, aiming for lower latency and parallel tool calls. ( ) - Anthropic published Constitutional 2.0 and a new Claude model, while the FAA announced a Palantir partnership for AI air-traffic control. ( ) - The flurry matches other multimodal moves like Gemini 2.5 Pro's editable 3D output and VideoLLaMA 3's SOTA video comprehension. ( )
Artificial intelligence companies spent the past week pushing the same promise from different angles: faster models, more tool use, and more work handed off to software. (openai.com) OpenAI said its o-series models are built to “think for longer” before answering, and said o3 and o4-mini can use ChatGPT tools including web search, Python, file analysis, vision, and image generation. OpenAI’s developer docs also say function calling lets models trigger outside software through JSON-defined tools, with parallel tool calling available in supported setups. (openai.com) (developers.openai.com) Function calling is the plumbing behind many “agent” demos: the model does not just write text, it asks an app to fetch data, run code, or take an action, then uses the result in its reply. OpenAI’s docs describe that as a multi-step loop between the model and the developer’s application, and its older Assistants docs say newer models can invoke parallel function calls for lower latency on independent tasks. (developers.openai.com 1) (developers.openai.com 2) Anthropic moved on a parallel track. On January 22, 2026, it published a new constitution for Claude under a Creative Commons CC0 license, saying the document directly shapes Claude’s behavior during training by spelling out values, tradeoffs, and instructions for difficult situations. (anthropic.com) Then on April 16, 2026, Anthropic released Claude Opus 4.7, saying the model improves on Opus 4.6 in advanced software engineering, handles long-running tasks more consistently, and is available across Claude products, the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. Anthropic listed pricing at $5 per million input tokens and $25 per million output tokens, unchanged from Opus 4.6. (anthropic.com) The Federal Aviation Administration is making a separate bet on automation in a higher-stakes setting. The agency says it is replacing core air-traffic infrastructure including radar, software, hardware, and telecommunications networks, with a goal of implementing a new system by the end of 2028. (faa.gov) That effort is already being parceled out through large contracts. In December 2025, the Federal Aviation Administration said Peraton would serve as prime integrator for the broader overhaul, while Palantir has an older Federal Aviation Administration contract tied to aircraft-certification safety monitoring for the 737 Max fleet rather than the main air-traffic-control rebuild. (faa.gov) (palantir.com) Google and open-source researchers are pushing the same multimodal direction. Google said in May 2025 that Gemini 2.5 Pro reached state-of-the-art results on key video-understanding benchmarks and showed how the model could turn a YouTube video into an app spec and then into code. (developers.googleblog.com) Across these releases, the common shift is from chatbots that answer questions to systems that inspect images, watch video, call tools, and return work products. The race is no longer just over who writes the best paragraph, but who can complete the next step without waiting for a human prompt. (openai.com) (anthropic.com) (developers.googleblog.com)
