|
|
|
|
|
|
At $2/M input through August 31, effort-scaled Sonnet 5 matches Opus 4.8 at a fraction of the cost, per Anthropic. The price jumps to $3/M on September 1.
|
|
JULY 4TH, 2026 · BY JONATHAN HILDEBRANDT
|
|
Anthropic shipped Claude Sonnet 5 on June 30 and made it the default for Free and Pro users. Through August 31 it runs at $2/M input and $10/M output, then jumps to $3/$15 on September 1, per Anthropic. At higher effort levels Anthropic says it matches Opus 4.8 (which is $5/$25) on BrowseComp and OSWorld-Verified, meaning 2.5 Sonnet 5 input calls cost what one Opus 4.8 input call does. Why it matters. If your agent stack pins Opus 4.8 for hard tasks, the effort-scaled Sonnet 5 is worth a re-benchmark before the intro price expires on August 31, per the Anthropic announcement.
|
|
Read our Sonnet 5 writeup →
|
|
|
|
•
|
Anthropic will pay up to $15,000 per jailbreak report. The new HackerOne program pairs the bounty with a CVSS-style severity rubric, the first lab to score jailbreak severity on a reproducible scale. Details →
|
|
•
|
Cloudflare split AI crawlers into three lanes you can price separately. Search, Agent, and Training each get their own allow/block/charge control. On Sept 15 new unmanaged zones default to blocking Training, a GEO decision for every content site. Details →
|
|
•
|
Make: hosted automation with AI agents, no infrastructure to run. Watch the credit math: Make bills one credit per operation plus token-based credits on AI steps (1 credit per 723 GPT-5 input tokens, 1 per 90 output). Try Make →
|
|
•
|
CustomGPT: no-code GPT trained on your own knowledge base, SOC 2 and GDPR covered. Useful for support chat over your docs without building retrieval yourself. Try CustomGPT →
|
|
|
|
|
Jonathan Hildebrandt
Co-founder and primary operator of Pondero. Writes the Pondero Brief.
|
|
|
Affiliate disclosure
·
Pondero earns commissions on some links. This does not affect our editorial picks.
|
|
