Claude Opus 5 Arrives — Near-Fable Performance at Half the Price
Anthropic launched Claude Opus 5, its new flagship for coding, agents, and complex knowledge work. The headline claim: it approaches — and on several benchmarks surpasses — the pricier Fable 5, while costing roughly half per task. It ships with a 1M-token context window, up to 128K tokens of synchronous output, and a new five-level effort control that lets you dial how hard the model works against how fast it responds.
Opus 5 is available across every Anthropic platform and is already the default on Claude Max and the strongest option on Claude Pro. Early testers highlighted its knack for verifying its own work and iterating until it succeeds — in one viral demo it built a computer-vision pipeline from scratch to reconstruct a machine part as a 3D model. This is the model the rest of today’s news orbits.
Voice Mode Expands to Opus and Sonnet, Plus Connected Tools
Claude’s voice mode got a major upgrade: it now runs on Opus and Sonnet, supports connected tool actions, and adds many more languages. You can talk through real work, switch models mid-conversation, and move from a spoken conversation straight into doing — across mobile, desktop, and web.
The shift matters because voice stops being a novelty demo and starts being a way to actually drive Claude. Ask a question out loud, have it call a connected tool, and get work done without touching the keyboard — the same agentic muscle behind Cowork and Claude Code, now available hands-free.
Cowork Expands Across Mobile and Web
Claude Cowork — the agent that works directly with your files and runs tasks in the background — is widening its rollout so sessions and files follow you across devices. Start work at your desk, approve a step from your phone, and pick up scheduled tasks, shared chats, and projects wherever you are.
The pattern is consistent with Anthropic’s read of how people actually use Cowork: the vast majority of sessions aren’t coding — they’re business operations and content work. Making the agent portable, with background execution and mobile approvals, is how Anthropic pushes that everyday-office use case further into daily habit.
Claude Code Ships Opus 5 as Default, Adds Nested Subagents
Claude Code now runs Claude Opus 5 as its default Opus model and layers on dynamic workflows, nested subagents, improved MCP handling, and more reliable runners. It also tightens the screws on safety: stricter sandbox controls, better error reporting, and fixes across model selection, remote control, and terminal behavior.
Nested subagents are the standout — an agent can now spin up its own sub-agents for a task, which is the kind of hierarchy real multi-step engineering work demands. Paired with Opus 5’s 1M context, Claude Code can hold far more of a codebase in view while it works.
Opus 5 API Pricing: $5/$25, With a Faster Mode
For builders, Opus 5 lands at $5 per million input tokens and $25 per million output — the same API pricing as Opus 4.8, for a stronger model. There’s also a fast mode that runs about 2.5x quicker than the default at twice the cost (roughly $10 / $50 per Mtok), for latency-sensitive workloads.
The economics are the real story: the same price as the model it replaces, near-Fable capability, and a five-level effort dial so you only pay for the compute a task actually needs. For agent builders watching token spend, that’s a meaningful upgrade without a bill shock.
Anthropic Economic Index Comes to Claude as a Connector
Anthropic shipped the Anthropic Economic Index connector, letting anyone explore AI-usage data directly inside a Claude chat. Ask about occupations, tasks, and adoption trends, and Claude answers grounded in the Index — with the underlying source data available right in the conversation.
It’s a neat demonstration of the connector model: instead of downloading a dataset and wrangling it yourself, you query it conversationally. For researchers, journalists, and policy folks tracking how AI is reshaping work, it turns a static report into something you can actually interrogate.
Claude Code and Cowork Reach Government in Public Beta
Anthropic put Claude Code and Claude Cowork into public beta inside Claude for Government Desktop, delivered in a FedRAMP High authorized environment and spanning all three branches of the federal government. Agency staff can delegate memos, RFP reviews, casework, and decks to Cowork, or modernize software systems with Claude Code.
Governance is the selling point: admins set configuration defaults and control spending across departments, while security teams get tamper-evident audit logs and ATO documentation. A limited-time program even makes it available to agencies at $1 per agency (unlimited seats) through August 2026.
The Opus 5 Playbook: Push Capability Down, Push Price Down With It
Yesterday was plumbing; today is the payload. Opus 5 isn’t just faster — it’s Fable-class work at Opus-class prices, and Anthropic wired the whole stack to absorb it in a day. Claude Code makes it default, voice mode carries it hands-free, Cowork carries it across your devices, and government buyers get it under FedRAMP High. That’s a launch designed to land everywhere at once.
The quiet strategic move is the five-level effort control. Anthropic is telling builders: stop choosing between models on capability, and start choosing on how much compute each task deserves. When the frontier gets cheaper and the dial gets finer, the question shifts from “can the model do this?” to “how little am I willing to spend to have it done well?” That’s the sound of frontier AI becoming a metered utility — and yesterday’s boring infrastructure work is exactly what makes a day like today possible.