Anthropic Claude Sonnet 5 announcement

TL;DR

Claude Sonnet 5 is Anthropic’s newest Sonnet‑class model that delivers near‑Opus‑class agentic capabilities—planning, tool use, coding, and knowledge work—at a lower price point and with improved safety.

What makes Sonnet 5 more agentic?

Sonnet 5 is described as the "most agentic Sonnet model yet." It can autonomously create plans, invoke external tools such as browsers and terminals, and execute multi‑step tasks that previously required larger, more expensive models. Compared with its predecessor Sonnet 4.6, Sonnet 5 shows measurable gains in reasoning, tool use, coding, and knowledge‑work performance.

Performance relative to other Anthropic models

  • Close to Opus 4.8: Benchmark tables in the announcement show Sonnet 5’s scores on a variety of evaluations are comparable to Opus 4.8, a higher‑tier model, while retaining Sonnet‑class pricing.
  • Improvement over Sonnet 4.6: Across agentic search (BrowseComp) and computer‑use (OSWorld‑Verified) tasks, Sonnet 5 consistently outperforms Sonnet 4.6 and offers a broader cost‑performance envelope than Opus 4.8.
  • Cost‑efficiency: At $2 per million input tokens and $10 per million output tokens, Sonnet 5 delivers higher performance per dollar, especially at medium effort levels where it can match Opus 4.8 on some tasks.

Real‑world developer feedback

Early‑access partners report that Sonnet 5 completes complex, multi‑step workflows that earlier Sonnet models abandoned. Representative quotes include:

"Claude Sonnet 5 gives our agents a strong execution layer for multi‑step software engineering work. It handles sustained coding, tool use, and debugging well across messy technical contexts."

"We handed Claude Sonnet 5 a two‑part job—update Salesforce account tiers, send a launch announcement to enterprise contacts—and it finished end to end. That used to stall halfway."

"Claude Sonnet 5 gets more done with less. Same output quality, fewer steps to get there. It refuses unsafe requests cleanly and consistently."

"I asked Claude Sonnet 5 to investigate a bug. Unprompted, it wrote a reproducing test, implemented the fix, then stashed it to confirm the bug came back without the change. All in a single pass."

These anecdotes illustrate the model’s ability to maintain focus, follow conventions, and produce verified code changes without constant prompting.

Safety profile

Anthropic’s pre‑deployment safety evaluations indicate:

  • Lower overall undesirable behavior than Sonnet 4.6, including fewer hallucinations, sycophancy, and better refusal of malicious requests.
  • Higher safety than Opus 4.8 on agentic misuse, though Sonnet 5 still shows a slightly higher misalignment rate than the Mythos preview and Opus 4.8 on a broad automated behavioral audit.
  • Reduced cybersecurity capability: Sonnet 5 was never able to develop a full exploit in a Firefox‑based test, and its partial‑success rate remains low. Nonetheless, it is marginally higher than Sonnet 4.6, likely due to general intelligence gains.

Because of the modest increase in cyber‑related abilities, Anthropic ships Sonnet 5 with real‑time cyber safeguards enabled by default—identical to those used for Opus 4.7/4.8 but less restrictive than those for Fable 5.

Availability and pricing

  • General availability: Sonnet 5 is the default model for Free and Pro plans and is also available to Max, Team, and Enterprise customers.
  • Pricing: $2 per million input tokens and $10 per million output tokens (permanent as of August 10 2026). Rate limits have been raised across Claude Chat, Cowork, Claude Code, and the Claude Platform to support higher‑effort usage.
  • API access: Developers can invoke the model via the Claude API using the identifier claude-sonnet-5.

Implications for the AI ecosystem

  • Agentic AI democratization: By narrowing the performance gap between Sonnet and Opus families, Anthropic makes high‑quality autonomous agents more affordable for a broader developer base.
  • Competitive pressure: The cost‑performance frontier established by Sonnet 5 may push other providers to release similarly capable, lower‑cost agentic models.
  • Safety trade‑offs: While safety has improved relative to the previous Sonnet, the model remains less safe than the top‑tier Opus models, highlighting the ongoing balance between capability and alignment.

All figures, charts, and direct quotations are taken from Anthropic’s official Claude Sonnet 5 announcement and system card.

Sources

Related