🧬 Anthropic’s Claude Science bets on workflow, not a new model, to win over scientists ↗
Anthropic introduced Claude Science, an AI workbench for computational research that pulls databases, pipelines, tools, and project history into one place. Not a shiny new model, surprisingly enough - more like a lab bench with an over-caffeinated assistant. (TechCrunch)
The setup connects to 60+ scientific databases, supports toolkits for genomics, protein structure, and chemistry, and includes a fact-checking AI for citations and calculations. Handy, though faintly fox-guarding-the-henhouse, since it is still AI checking AI.
Early users at the Allen Institute and UCSF are already using it for review pipelines and glioma analysis workflows. Anthropic’s bigger play seems fairly clear: own the workflow layer, not just the model race.
🤖 Anthropic launches Claude Sonnet 5 as a cheaper way to run agents ↗
Anthropic also rolled out Claude Sonnet 5, pitching it as a cheaper, more agentic midsize model that can plan, use tools, and run tasks more autonomously. The “agent” drumbeat is getting loud now. (TechCrunch)
It starts as the default for free and Pro users, with pricing at $2 per million input tokens and $10 per million output tokens before rising later. That puts it below Opus 4.8, GPT-5.5, and Gemini 3.1 Pro on cost, though not below Gemini 3.5 Flash.
The safety angle matters too. Anthropic says Sonnet 5 shows fewer unwanted behaviours than Sonnet 4.6, with better resistance to prompt-injection attacks - not perfect, but less wobbly on the bicycle.
🍌 Google introduces a faster, cheaper image generator with Nano Banana 2 Lite ↗
Google released Nano Banana 2 Lite, a lower-latency version of its AI image and video generator aimed at high-volume creative workflows. The name is silly, the strategy is not. (TechCrunch)
Google says it can generate images in around four seconds and costs $0.034 per 1,000 images. That is clearly built for people churning through drafts, ad concepts, product shots, and the whole content sausage machine.
It is available via Google AI Studio, the Gemini API, and Gemini Enterprise Agent Platform. Google also widened Gemini Omni Flash and showed Omni Product Studio, which turns static images into e-commerce style videos.
🛡️ Lumo, Proton’s privacy-focused AI chatbot, gets an upgrade ↗
Proton upgraded Lumo to version 2.0, adding image recognition, image generation, Projects support, persistent memory controlled by users, and a new thinking mode. Not bad for a privacy-first chatbot trying to compete in a very noisy alley. (TechCrunch)
The company says responses are up to 76% faster than before. The bigger sell, though, is privacy: Proton says chats use zero-access encryption, no server-side session logging, no training on customer data, and no third-party sharing.
So Lumo is chasing the same general usefulness as ChatGPT or Gemini, but with a sharper privacy pitch. Whether that wins mainstream users is the crunchy bit.
💸 Crypto exchange OKX wants AI agents to hire and pay each other ↗
OKX launched OKX AI, a marketplace where AI agents can hire each other, settle payments, and build portable on-chain reputations. It sounds slightly sci-fi, slightly spreadsheet, and very venture-backed. (TechCrunch)
The marketplace opens to developers after a closed beta with 50 early AI service providers. OKX says the system builds on agent wallets, stablecoin payments, and persistent identities.
Early builders include CertiK for wallet and token security checks, CoinAnk for market data, and GenLayer for dispute resolution. That last part is peculiarly important - autonomous agents need a tiny digital court, apparently.
🧯 US removes curbs on Anthropic's latest Fable and Mythos AI models ↗
The U.S. Commerce Department lifted export controls on Anthropic’s Fable 5 and Mythos 5 models after new safeguards were put in place. That ends a tangled restriction period that had forced Anthropic to cut off broad access. (Reuters)
Mythos 5 had been tied to cybersecurity vulnerability detection, while Fable 5 is intended for broader public use with stronger safeguards. Anthropic says some benign requests may now get blocked and routed elsewhere, which is annoying, but probably the compromise.
The bigger story is government getting deeper into frontier model release decisions. OpenAI has faced similar scrutiny, so this is not just one company’s headache - it is the new weather system.
FAQ
What is Claude Science, and why is Anthropic focusing on workflow?
Claude Science is described as an AI workbench for computational research, rather than a new standalone model. It brings scientific databases, pipelines, tools, and project history into a single environment. The goal appears to be helping researchers manage complex workflows more fluidly, especially in areas such as genomics, protein structure, chemistry, review pipelines, and glioma analysis.
How does Claude Science support scientific research workflows?
Claude Science connects to more than 60 scientific databases and supports specialist toolkits for genomics, protein structure, and chemistry. It also includes an AI fact-checking layer for citations and calculations. In practice, this allows researchers to keep more of their literature review, data access, analysis steps, and project context inside one shared workspace.
What makes Claude Sonnet 5 valuable for AI agents?
Claude Sonnet 5 is positioned as a cheaper midsize model for agentic work. It can plan, use tools, and run tasks with more autonomy than a basic chatbot-style setup. Anthropic is also pitching improved safety behaviour, including stronger resistance to prompt-injection attacks compared with Sonnet 4.6, which matters when agents interact with external tools and data.
Why are cheaper AI models important for production workflows?
Cheaper models can make high-volume AI use more practical, especially when teams are running agents, generating drafts, or processing large numbers of requests. Claude Sonnet 5 and Nano Banana 2 Lite both reflect this direction: lower-cost systems aimed at frequent usage rather than only premium, one-off tasks. For businesses, cost per token or output can shape what is feasible to deploy.
How is Nano Banana 2 Lite different from other AI image generators?
Nano Banana 2 Lite is designed for faster, lower-cost image and video generation. Google says it can create images in around four seconds and is priced for high-volume creative workflows. That makes it especially relevant for teams producing ad concepts, product shots, content drafts, or e-commerce visuals, where speed and iteration matter as much as polish.
What is the privacy pitch behind Proton’s Lumo 2.0?
Lumo 2.0 is Proton’s privacy-focused AI chatbot, now upgraded with image recognition, image generation, Projects, user-controlled memory, and a thinking mode. Its main pitch is privacy: Proton says chats use zero-access encryption, are not logged server-side by session, are not used for customer-data training, and are not shared with third parties.