Claude Sonnet 5 is now the default model for Free and Pro users
If you use Claude in Chat, Cowork, Claude Code or via the API, Sonnet 5 is now the model doing the work by default for Free and Pro users, with Max, Team and Enterprise able to select it too. Anthropic says it closes much of the gap to Opus 4.8 on agentic tasks including multi-step coding, browser use and long tool-calling chains, while costing far less to run. Early testers describe it finishing jobs that previously stalled halfway through.
Introductory pricing is $2 per million input tokens and $10 per million output tokens until 31 August 2026, after which it rises to $3 and $15. For anyone building agents on a budget, this is the most practical upgrade of the week: broadly similar agentic capability to the flagship model at a fraction of the cost. Cybersecurity safeguards are switched on by default, though Anthropic notes Sonnet 5's ability to develop working exploits remains well below that of its Opus models.
Claude Science opens in beta with 60-plus scientific database connectors
Anthropic has opened a beta of Claude Science, a version of its assistant built for lab work rather than general chat. It comes pre-loaded with connectors for more than sixty scientific databases spanning genomics, proteomics, single-cell analysis and cheminformatics, and produces auditable artefacts so a researcher can trace how a figure or conclusion was reached. It runs on the same Claude Opus 4.8 already available to everyone, wrapped in tools built for scientific workflows.
Anthropic is also backing up to fifty research projects with credits worth up to $30,000 each, with applications open until 15 July 2026. If your work touches biology, chemistry or any data-heavy research field, this is worth investigating even if you have never used Claude for coding or writing. The same-day launch of Anthropic's Claude Science and OpenAI's GeneBench-Pro marks biology as the latest competitive front between the two labs.
GeneBench-Pro is a harder successor to the genomics benchmark, and even the best model solves fewer than a third of it
OpenAI has introduced GeneBench-Pro, a tougher successor to its original genomics benchmark that tests whether a model can make the kind of messy judgement calls real computational biologists face, rather than simply processing well-formatted inputs. The company says its strongest model, GPT-5.6 Sol, solves under a third of the 129 test problems even at its highest reasoning setting, up from below five per cent for GPT-5 on the original benchmark.
It matters less for casual users and more as a signpost: expect biology and drug discovery to keep appearing as a proving ground for frontier models over the coming months, especially now that Anthropic is pushing Claude Science on the very same day. The two labs are converging on science as the next credibility test alongside coding.
From today, Microsoft 365 Copilot Business rolls off its $18 promotional rate to a standard $21 per user per month. Two new bundled plans, Business Standard and Business Premium with Copilot, go live for new commercial customers. Nothing new to try, but worth flagging for renewal planning.
AWS has stood up a new internal team of forward-deployed engineers who embed directly with customers to build bespoke AI agents, following similar moves at OpenAI and Anthropic. It signals Amazon wants Bedrock customers getting hands-on help turning pilots into production rather than just API access. TechCrunch, 30 Jun →
OpenClaw is now available on iOS and Android. The free open-source agent pairs your phone with the OpenClaw Gateway to route requests to agents and skills for multi-step tasks. The easiest way in to open-source agents for anyone who has not wanted to set one up on a laptop. Proton Lumo 2.0 adds image recognition and generation, closing a feature gap with mainstream assistants for privacy-first users. TechCrunch, 30 Jun →
Today's themes
Two patterns stand out today. First, the agentic push keeps getting cheaper and more accessible: Sonnet 5 delivers near-flagship agentic performance at a fraction of Opus pricing, and OpenClaw's mobile launch brings a free open-source agent to phones for the first time, both pointing towards agents becoming a commodity rather than a premium feature. Second, science is emerging as a competitive battleground beyond coding: Anthropic's Claude Science and OpenAI's GeneBench-Pro landed on the same day, each staking a claim on biology and research workflows. For teams in life sciences or data-heavy research, this is the week to start evaluating these tools seriously.