|
|
|
Claude Sonnet 5 is now the default. The Aug 31 pricing cutoff is the date to watch.
Sonnet 5 approaches Opus 4.8 on agentic tasks at introductory $2/M input pricing, through Aug 31, per Anthropic.
See the pricing and effort curves →
|
|
|
|
Sonnet 5 approaches Opus 4.8 on agentic tasks at introductory $2/M input pricing, through Aug 31, per Anthropic.
|
|
JULY 1ST, 2026 · BY JONATHAN HILDEBRANDT
|
|
Anthropic shipped Claude Sonnet 5 on June 30 and made it the default for Free and Pro plans the same day. It approaches Opus 4.8 on BrowseComp (agentic search) and OSWorld-Verified (computer use) at higher effort levels. API id is claude-sonnet-5. Through Aug 31 it runs $2 per million input tokens and $10 output; on Sep 1 it moves to $3 and $15, per Anthropic's announcement.
Why it matters. Standard-rate Sonnet 5 lands at 60 percent of Opus 4.8's $5/$25 (Anthropic), and the intro rate is cheaper still, so the buy question is which agent jobs you can demote from Opus before Aug 31. One catch: the new tokenizer maps the same input to roughly 1.0 to 1.35 times more tokens, which Anthropic prices as cost-neutral. It is live in Claude Code now, with model selectors in Cursor and Cline.
|
|
See the pricing and effort curves →
|
|
|
Tool to try: Make
PONDERO PICK
Connect an LLM call to a CRM update, a Slack post, or a webhook, no backend required. Make is the cleanest path from an AI model output to something that actually happens. Try Make →
|
Tool to try: Cloudways
PONDERO PICK
Managed cloud hosting with server-in-minutes setup. When you want your own AI stack instead of a metered API, Cloudways is the least-friction host for Open WebUI, n8n, and similar self-hosted tools. Try Cloudways →
|
|
•
|
Microsoft flagged MCP tool poisoning as a live threat. A poisoned tool description can make M365 Copilot, Copilot Studio, and Azure AI Foundry silently hand over data with no security alert. Audit your tool descriptions as untrusted input this week. Details →
|
|
•
|
Apple credited AI tools by name in a security advisory for the first time. iOS 26.5.2 and macOS Tahoe 26.5.2 patch 30-plus vulnerabilities; four WebKit CVEs were found by OpenAI Codex Security and Anthropic researchers using Claude. AI bug-hunting just crossed from vendor PR into a shipping Apple advisory. Details →
|
|
•
|
GitHub Copilot's token billing is producing sticker shock. Teams on heavy agentic usage are seeing $750 to $3,000 monthly against the old $29 to $50 flat plans. Multi-step edits and long context are the primary drivers. Meter your agent-mode usage before month-end. Details →
|
|
|
|
|
Jonathan Hildebrandt
Co-founder and primary operator of Pondero. Writes the Pondero Brief.
|
|
|
X
·
LinkedIn
·
Bluesky
·
Affiliate disclosure
·
Pondero earns commissions on some links. This does not affect our editorial picks.
|
|
