🧭 Claude Sonnet 5: A 1M-Token Context Window at Sonnet Pricing, Beating Opus on Terminal-Bench

Anthropic launched Claude Sonnet 5 (model ID claude-sonnet-5), calling it its most agentic Sonnet yet. The headline spec: a native 1M-token context window — the first time Anthropic's mid-tier model has shipped frontier-scale context as standard, rather than context depth being reserved for the Opus tier. Sonnet 5 plans multi-step work, drives browsers and terminals, and runs autonomously at a level Anthropic says recently required larger, more expensive models.

Pricing and availability

• Introductory pricing: $2 per million input tokens / $10 per million output tokens through August 31, 2026 — then standard pricing of $3/$15 per million tokens.
• Live everywhere at once: Claude Code, the Claude API, claude.ai, Cursor, VS Code, and GitHub Copilot.
• On paper, costs the same as Sonnet 4.6 at introductory pricing, and well under half of Opus 4.8's rate.

Benchmark results

Sonnet 5 posts major gains over Sonnet 4.6 across every evaluation Anthropic disclosed, and narrows the gap to Opus 4.8 across five major evaluations — surpassing the flagship on one. It leads SWE-bench Pro at 63.2%, ahead of GPT-5.5 (58.6%) and Gemini 3.5 Flash (55.1%). GPT-5.5 edges it out on Terminal-Bench 2.1 (83.4% vs. 80.4%) — but Sonnet 5 is the first time a mid-tier Sonnet model has beaten its own Opus sibling on a major coding benchmark, directly challenging the assumption that terminal-agent work requires the larger model.

🧭 Claude Science Launches: 60+ Research Tools in One Workbench for Drug Discovery

Anthropic also launched Claude Science, an AI workbench that consolidates more than 60 scientific databases and computational tools — spanning genomics, structural biology, proteomics, and cheminformatics — into a single research environment. The pitch, in Dario Amodei's framing, is that Claude Science does for life sciences what Claude Code has done for programming: researchers normally juggling dozens of separate programs can give the platform a single instruction and have it carry out the analysis across tools that don't otherwise talk to each other.

Launch details

• Availability: beta, immediately available to existing Claude Pro, Max, Team, and Enterprise subscribers — no separate signup required.
• Industry validation: the launch event featured a panel including Bristol Myers Squibb CEO Chris Boerner, Novartis CEO Vas Narasimhan, Genentech EVP Aviv Regev, and GLP-1 drug inventor Lotte Knudsen.
• Research grants: Anthropic simultaneously opened a grant program offering up to $30,000 in compute credits per project for up to 50 research teams, applications open through July 15, 2026.
• Anthropic's own drug programme: alongside the product launch, Anthropic is starting an internal drug discovery program — a direct move into the pharma value chain rather than staying purely a tools vendor.