Claude Opus 5 ships at half Fable 5's price, shared chats exposed in Google Search, Managed Agents upgrades

Anthropic released Claude Opus 5 on July 24 — its fourth model in under two months, following Mythos 5, Fable 5, and Sonnet 5. Priced at $5 per million input tokens and $25 per million output tokens (same as Opus 4.8, half of Fable 5's $10/$50), Opus 5 delivers near-Fable 5 benchmark performance: 79.2% on SWE-Bench Pro (vs Fable 5's 80.0% and Opus 4.8's 69.2%), 43.3% on Frontier-Bench v0.1 agentic coding (more than doubling Opus 4.8), and 30.2% on ARC-AGI-3 (three times the next-best score). It tops Artificial Analysis' Intelligence Index and Agentic Index on launch day. Ships with a 1M-token context window (default and maximum), 128K max output, thinking on by default, and a May 2026 knowledge cutoff — the most current of any Claude model. Available as claude-opus-5 on the Claude API, Claude Code, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry, and claude.ai. It is the new default on Claude Max and the strongest model available on Claude Pro.

Opus 5 introduces a per-request effort setting: low, medium, high, and max. Low is faster and cheaper for routine work; high lets the model spend more reasoning time on hard problems. This is Anthropic's first model to expose the effort ladder at the API level — pass it in the model config or set it in Claude Code with /effort. Fast mode is also available as a research preview on the Claude API at $10/$50 per MTok (~2.5x default speed). Batch API pricing is $2.50/$12.50 per MTok; prompt cache read hits are $0.50 per MTok. Unlike Fable 5, Opus 5 supports zero data retention.

A Reddit post on July 26 revealed that hundreds of Claude shared conversations were publicly discoverable through Google Search. Users querying site:claude.ai/share could access conversations containing legal advice, engineering work, and personal discussions without needing direct links from the original owners. The root cause: claude.ai/share pages were not excluded from search engine indexing. Anthropic has since added a noindex directive to shared conversation pages, but existing indexed links remain live and publicly accessible. If you previously shared a Claude conversation, that link may still be discoverable in search engines. Review your shared links at claude.ai/settings.

The Claude Developer Platform shipped a significant Managed Agents upgrade: pass effort inside the model object when creating an agent to set its reasoning intensity at the agent level. Webhook coverage expanded to include environment events and memory store events. Session seeding lets you pre-populate a new agent session with an initial set of events — useful for injecting context or replaying a known state. Optional version checks on updates let you gate agent restarts on a specific Claude version. Event deltas for thread streams reduce payload size by sending only changed fields instead of full event objects.

Voice mode is now available in the Claude iOS and Android apps. Tap the sound wave icon in the mobile app to start a voice conversation with Claude. The feature uses the same underlying models as text conversations and supports switching between voice and text mid-conversation.

The Anthropic Economic Index — a dataset tracking how AI is affecting occupations, tasks, and labor markets — is now queryable through Claude directly. Ask Claude questions about the index data from the announcement page. Anthropic also published the research agenda for the Anthropic Economic Futures Research Fund this week, outlining the grant priorities for economists studying AI's effects on work and wages.

Claude Code shipped emoji shortcode autocomplete (type : to trigger), clearer warnings before writing to session transcript files, and tighter controls for subagents and token budgets. Background session handling and Windows stability received a round of fixes. Also fixed: a startup hang when the Claude in Chrome extension is enabled but Chrome is not running, a 300ms delay revealing async content in Settings tabs and diff views, and reopening a just-stopped background session from the agents view starting a blank session instead of resuming.

Anthropic committed another $20 million to Public First Action, a nonprofit focused on AI governance and policy. This follows earlier donations to the same organization and is part of Anthropic's broader commitment to funding independent policy and governance research on AI.

Anthropic opened applications for a new round of AI for Science grants focused specifically on rare disease research. The program funds researchers using Claude Science and related tools to accelerate drug discovery, biomarker identification, and patient stratification for diseases affecting fewer than 200,000 people in the US.