The Rundown AI is reporting that Anthropic has launched Sonnet 5, its latest mid-tier artificial intelligence model, which it describes as its "most agentic Sonnet model yet." This release comes as the U.S. Department of Commerce lifted export controls on Anthropic's more powerful Fable 5 and Mythos 5 models, which are set to return online. The newsletter said Sonnet 5 demonstrates significant advancements in agentic coding and reasoning compared to its predecessor, with knowledge work capabilities surpassing Opus 4.8. It noted the new Sonnet can operate a browser or terminal and handle longer tasks, bringing Opus-style agent behavior to Anthropic's more affordable tier. However, its cybersecurity benchmarks are lower than Sonnet 4.6, with Anthropic stating it "did not deliberately train" the model on cybersecurity tasks. Sonnet 5 is available across all plans, with API rates set at $2 per million input tokens and $10 per million output tokens until Aug. 31, increasing to $3 and $15 thereafter.
The Rundown AI also reported that Google introduced two new media models for developers: Nano Banana 2 Lite and Gemini Omni Flash. It said Nano Banana 2 Lite is a lower-cost option designed for fast, bulk image generation, capable of producing an image in four seconds at a cost of $.034 per image. The newsletter stated Gemini Omni Flash offers robust video generation and editing capabilities, creating and editing 10-second video clips at $.10 per second. It noted Omni Flash leads text-to-video leaderboards and is second only to Seedance 2.0 in video editing, integrating Gemini’s multimodal reasoning for real-world knowledge. Google aims to link these models, allowing users to create an image with Lite and then animate it into a video clip using Omni in a single workflow.
Additionally, The Rundown AI said Anthropic debuted Claude Science, an AI workspace for scientists that integrates paper review, databases, figure-making and computing jobs into a unified hub. It reported Anthropic is also initiating its own preclinical drug discovery program focused on "neglected" diseases. The platform is designed to track every step of the research process for auditing and fact-checking, connecting over 60 scientific sources and tools. Claude Science is available on paid plans for MacOS and Linux, with options for sensitive datasets to remain on local machines.
In other AI developments, The Rundown AI reported OpenAI discovered a "compute multiplier" that more than halves its inference costs, coinciding with the launch of its Jalapeño chip. U.S. chipmaker Etched announced $800 million in funding, a working inference chip, and $1 billion in customer contracts. AWS is committing $1 billion to a new Forward Deployed Engineering organization to embed engineers within customer teams to accelerate agentic AI builds. X rolled out a hosted MCP server, allowing AI tools like Grok, Claude, and Cursor to connect to the X API. OpenAI also teased a hardware collaboration with Work Louder, believed to be a keyboard device for its Codex platform, set to launch July 15.