English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

Claude Sonnet 5 Released: Anthropic's New Flagship-Adjacent Model at 60% of Opus Pricing

Forum topic · 小凯 · 2026-07-01

Summary

Anthropic released Claude Sonnet 5 on June 30, 2026, marking the first time the Sonnet series approaches Opus-class agentic capabilities. The model scores 63.2% on SWE-bench Pro and 81.2% on OSWorld-Verified, with native 1M-token context, and reportedly matches Opus 4.8 on some high-effort tasks. Pricing is $2/$10 per million tokens (input/output) until August 31, 2026, then $3/$15 — about 60% of Opus 4.8's $5/$25. Sonnet 5 is the first Sonnet model to include Anthropic's Cyber Verification Program, a default-on safeguard that detects and blocks dangerous cyber operations. It is available as the default model on Free/Pro plans, across Max/Team/Enterprise, and in Claude Code and the Claude Platform under API ID claude-sonnet-5. Notable caveats: a new tokenizer inflates token counts by 1.0-1.35x for the same text, cybersecurity capability trails Opus 4.8, and Anthropic corrected its cost-performance chart after launch due to a simplified BrowseComp methodology. Early production users include ClickHouse, Pace, and Lovable. The release reshapes AI coding economics, letting previously Opus-only agent pipelines run at substantially lower cost.

On June 30, 2026 at 18:02 UTC, Anthropic officially launched Claude Sonnet 5 — the first Sonnet model to approach Opus-class agentic capabilities. According to Anthropic, Sonnet 5 strictly outperforms Sonnet 4.6 across reasoning, tool use, coding, and knowledge work, and can even match Opus 4.8 on some tasks at high effort settings.

Pricing

  • Early-bird price (before Aug 31, 2026): $2 / million input tokens, $10 / million output tokens
  • Standard price (after Aug 31): $3 / million input tokens, $15 / million output tokens
  • For comparison, Opus 4.8 costs $5 / $25. Sonnet 5's standard price is 60% of Opus 4.8's, with a much narrower performance gap.

    Performance

    Reported figures (via aihot summary and aimadetools coverage):

  • SWE-bench Pro: 63.2%
  • OSWorld-Verified: 81.2%
  • Context: native 1M tokens
  • On BrowseComp (agentic search) and OSWorld-Verified (computer use), the cost-performance curve fully covers Sonnet 4.6 and approaches or surpasses Opus 4.8 in the mid-to-high effort range.
  • Sonnet 5 also joins Anthropic's Cyber Verification Program — real-time detection and blocking of dangerous cyber operations, enabled by default. This is the first time a Sonnet model ships with this protection layer.

    Availability: default model on Free / Pro, fully available on Max / Team / Enterprise, live in Claude Code and the Claude Platform. API ID: claude-sonnet-5.

    A note of caution: Anthropic corrected its cost-performance chart after launch because the initial BrowseComp data used a simplified method that underestimated Sonnet 5. With the standard method, the performance curve moved up overall.

    Why This Matters

    Sonnet 5 is not a minor tune-up of Sonnet 4.6 — it repositions Anthropic's workhorse model. For two years, Sonnet was the affordable middle-tier workhorse while Opus handled the hardest tasks. Sonnet 5 pushes that line to Opus's doorstep. As Anthropic's announcement puts it, tasks that "just a few months ago, required larger and more expensive models" can now run on Sonnet 5, 40% cheaper.

  • For developers: 63.2% SWE-bench Pro + 1M context + tool use + native agentic ability means agent pipelines that previously required Opus can now run on Sonnet 5. Default-on cybersecurity safeguards reduce compliance friction for enterprise deployment.
  • For competition: 90% of the performance at 60% of the price effectively eats into the Opus segment. OpenAI's o3-pro / o4, DeepSeek's R2, and Google's Gemini 2.5 Pro will need to reconsider pricing — Sonnet 5 sets a new baseline in agentic coding.
  • For Anthropic: token consumption per task will likely rise (high effort + long context), but the pre-Aug 31 pricing is intended to be cost-neutral. The bet: more usage at lower unit prices means higher total revenue.
  • Early customer signals (officially cited): ClickHouse uses it for live data exploration, Pace for insurance FNOL workflows, and Lovable values its ability to "refuse unsafe request cleanly." These are real production environments, not benchmark fabrications.

    Risks and Open Questions

  • New tokenizer: the same text maps to 1.0–1.35× more tokens, affecting billing. While the early-bird price is cost-neutral, users after Aug 31 need to recalculate.
  • Cybersecurity capability trails Opus 4.8 and Mythos 5. Anthropic explicitly says it did not specially train cyber tasks, so exploit-related scenarios still need higher-tier models.
  • The corrected chart is a reminder: wait a couple of days before trusting AI company launch-day benchmarks.
  • Anthropic is tightening cybersecurity oversight; default-on safeguards are self-restraint, but some legitimate research scenarios may require an enterprise unlock process.
  • Sonnet 5 is not a small upgrade — it redraws the capability ceiling of Anthropic's workhorse tier. AI coding toolchains will now be re-priced around Sonnet 5.

    References:

  • Official announcement: https://www.anthropic.com/news/claude-sonnet-5
  • System card: https://www.anthropic.com/claude-sonnet-5-system-card
  • Secondary coverage: https://news.qq.com/rain/a/20260701A01X7000

Tags

#claude-sonnet-5#anthropic#ai-coding#llm-pricing#agentic-ai#swe-bench#model-release

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178208347