Anthropic Launches Claude Sonnet 5: Agentic Performance, New Tokenizer, and Per-Task Cost Surprises
Synthesis history
4 versions, newest first.
-
Version 4 2026-07-07 08:17 UTC · 227 items
New items this pass are primarily amplification of the per-task cost critique, now reaching mainstream outlets and Reddit, with no new substantive data or voices. One minor addition: item 39954 references Anthropic's in…
-
Version 3 2026-07-03 18:40 UTC · 219 items
This pass adds two genuinely new elements: a direct factual dispute over whether "Fable 5" is a real Anthropic model (item 39239), which becomes a new tension and open question; and additional benchmark data placing Son…
-
Version 2 2026-07-02 08:28 UTC · 191 items
Three new developments entered this pass: Anthropic lifted US export controls on Fable 5, returning it globally on July 1, and simultaneously launched Claude Science and Claude Desktop for Linux, placing the Sonnet 5 re…
-
Version 1 2026-07-01 02:23 UTC · 125 items
Anthropic launched Claude Sonnet 5 on June 30, 2026, positioning it as its most agentic Sonnet model to date, with benchmark scores approaching but not matching Opus 4.8 — 63.2% on SWE-bench Pro versus Opus 4.8's 69.2% …