Claude Sonnet 5: Anthropic's Mid-Tier Model Now Punches at Opus 4.8 Weight for Half the Cost

Six points on an agentic coding benchmark. That's all that separates Anthropic's brand-new Claude Sonnet 5 from its flagship Opus 4.8 on SWE-bench Pro (63.2% vs 69.2%). Yet Sonnet 5 costs roughly 60% less per token. For most teams, that math settles the debate before it starts.

Claude Sonnet 5 delivers near-Opus quality at Sonnet prices

Anthropic launched the model today, June 30, with introductory pricing of $2 per million input tokens and $10 per million output tokens through August 31. After that, standard rates kick in at $3/$15. Opus 4.8 still sits at $5/$25.

On the knowledge-work benchmark GDPval-AA v2, Sonnet 5 actually edges past Opus 4.8: 1,618 vs 1,615. On Humanity's Last Exam with tools, the gap is almost invisible: 57.4% vs 57.9%. That's flagship territory at mid-tier prices.

Sonnet 5's real edge: agentic reliability

Benchmarks tell part of the story. Early testers describe a model that finishes complex multi-step tasks where previous Sonnets would stop short, and checks its own output without being asked. Sonnet 5 is now the default model for Free and Pro plans, and also ships in Claude Code and on AWS Bedrock.

Safety metrics improved too. Lower hallucination rates, less sycophancy, and better resistance to prompt injection than Sonnet 4.6. Anthropic notes the model's cyber-offensive capabilities remain far below Opus 4.8, which keeps the risk profile manageable.

Claude Science: Anthropic's big bet on labs

The same day, Anthropic also unveiled Claude Science in beta: a desktop workbench for researchers with 60+ scientific databases, Nvidia's BioNeMo toolkit, and a reviewer agent that audits citations and calculations. Novo Nordisk, AstraZeneca, and Eli Lilly are already using it.

This tracks with Anthropic's broader strategy: build vertical products on top of its models, not just sell raw API tokens. Claude Code rewired how developers work; Claude Science aims to do the same for biology and chemistry.

What Claude Sonnet 5 means for the AI market

The playbook is clear. Send most agentic coding, tool use, and knowledge work to Sonnet 5. Reserve Opus 4.8 for accuracy-critical tasks. Keep Haiku 4.5 for high-volume, low-latency calls.

Against OpenAI's GPT-5.5 and Google's Gemini 3.1 Pro, Anthropic is making an aggressive move: near-frontier power, free to every user, months before a closely watched IPO. If you're building agents on a budget, today just became a very good day.