Anthropic dropped Claude Opus 4.7 with a 13% coding boost and megapixel vision, then in the same 48 hours quadrupled its London footprint right after OpenAI claimed its own permanent UK hub, turning the Knowledge Quarter into ground zero for the next phase of the AI arms race. Meanwhile Mozilla launched an open-source AI client that runs on your own hardware, Manycore Tech's spatial-intelligence IPO popped 187% on day one, and the US General Services Administration unveiled plans to let AI claw back a million work hours after losing 40% of its staff.
Anthropic Releases Claude Opus 4.7 with 13% Coding Gains and 3.75-Megapixel Vision
Anthropic launched Claude Opus 4.7 across all Claude products, its API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. The model delivers a 13% improvement over Opus 4.6 on a 93-task coding benchmark with particular gains on the hardest software engineering problems, achieves state-of-the-art scores on economically valuable knowledge work (GDPval-AA), and introduces substantially higher-resolution vision supporting images up to 3.75 megapixels. Pricing stays flat at $5/$25 per million input/output tokens, but a new tokenizer means inputs map to 1.0-1.35x more tokens. The release also includes a new 'xhigh' effort level for reasoning-intensive tasks and task budgets in public beta, signaling Anthropic's push toward more controllable, autonomous agent workflows.
Anthropic Unveils 800-Person London Office Days After OpenAI Claims First Permanent UK Hub
Anthropic announced plans to move into a 158,000-square-foot London office in the Knowledge Quarter with space for 800 staff, a quadrupling of its current 200-person UK headcount. The move arrived just days after OpenAI said it would open its own first permanent London office with room for more than 500 employees. Both companies are planting deeper flags in a single London district that already houses Google DeepMind, Meta, Synthesia, and Wayve, as the UK government courts Anthropic with a potential dual listing amid CEO Dario Amodei's planned May visit. The back-to-back expansions turn central London into the most concentrated AI hiring battleground outside the Bay Area.
Mozilla Launches Thunderbolt, an Open-Source Self-Hosted AI Client to Challenge Copilot and ChatGPT Enterprise
MZLA Technologies, the for-profit arm behind Thunderbird, released Thunderbolt: an open-source, self-hostable enterprise AI client built for organizations that refuse to pipe their data through Microsoft Copilot, ChatGPT Enterprise, or Claude Enterprise. Out of the box it supports Anthropic, OpenAI, Mistral, and OpenRouter as cloud providers, and runs local models via Ollama, llama.cpp, or any OpenAI-compatible endpoint. Native builds ship for Linux, macOS, Windows, iOS, and Android, with a web app alongside. The project, developed in partnership with Berlin-based Haystack maker deepset, is live on GitHub, currently under security audit, and explicitly pitched as a sovereign AI workspace with no vendor lock-in.
Manycore Tech IPO Soars 187% on Day One as 'Physical AI' Goes Public in Hong Kong
Shares of Hangzhou-based Manycore Tech jumped up to 187% in their Hong Kong debut after the spatial-intelligence firm raised about HK$1.02 billion ($156 million) in its IPO. Manycore is the first of the so-called 'Hangzhou Six Little Dragons' to list publicly, joining peers like DeepSeek, Unitree Robotics, and Game Science in a cohort that has captivated Chinese tech investors. The company's pitch: sell training data, SpatialLM models, and its SpatialVerse simulation platform to robot makers like partner Hesai, betting that the next frontier of AI is bridging language models into the physical world. About 30% of IPO proceeds will fund global expansion, with another 20% earmarked for new products targeting real-world and virtual environments.
GSA Launches 'Million Hours Challenge' for Internal AI Tool After Losing 40% of Workforce
The US General Services Administration unveiled plans to automate one million work hours of federal labor using USAi, its in-house generative AI tool, after shedding nearly 40% of its workforce since October 2024. Deputy Director Michael Lynch said the agency has already identified roughly 400,000 hours of automatable work, pushing it nearly halfway toward its goal, which is equivalent to a year of labor from 500 full-time employees. GSA is running an explicit 'eliminate, optimize, automate' playbook and now reports that almost half of its remaining workforce uses USAi daily. The initiative is the clearest sign yet that AI is being used to fill the void left by DOGE-era cuts across the federal government.
