Claude Sonnet 5 Launches: Near-Opus Performance at a Fraction of the Cost

The Story: The Most Agentic Sonnet Yet

On June 30, 2026, Anthropic launched Claude Sonnet 5, a Sonnet model built specifically for agentic workflows — autonomous tool use, planning, and multi-step reasoning.

The headline: Sonnet 5's performance approaches Opus 4.8 on many tasks, at a fraction of the cost. That's a real shift from earlier Sonnet models, which always traded off performance against price.

What Improved Over Sonnet 4.6

Compared to Sonnet 4.6, Sonnet 5 shows substantial gains in reasoning, tool use, coding, and knowledge work. On two benchmarks — BrowseComp (agentic search) and OSWorld-Verified (computer use) — Sonnet 5's cost-performance curve is markedly better: at the same cost, it gets more done.

In "higher-effort" mode, some Sonnet 5 tasks match Opus 4.8. Early testers report that Sonnet 5 completes multi-step tasks where previous models "would stop short," performs self-verification without being prompted, and shows better follow-through on software engineering work.

Pricing and a New Tokenizer

Anthropic is offering introductory pricing for Sonnet 5 through August 31, 2026: $2 per million input tokens, $10 per million output tokens. After that date, standard pricing kicks in: $3 per million input tokens, $15 per million output tokens — still well below Opus.

One detail easy to miss: Sonnet 5 uses a new tokenizer that consumes 1.0 to 1.35 times more tokens than the previous model for the same text. With AI token costs already a growing concern for many teams, this is worth factoring into any budget estimate before migrating.

On Safety

Anthropic reports that Sonnet 5 has lower rates of misaligned behavior than Sonnet 4.6, refuses malicious requests more reliably, and resists prompt injection attacks better. Hallucination and sycophancy rates are also lower.

One caveat: Sonnet 5's cybersecurity capabilities are substantially weaker than the Opus models. Anthropic ships the model with cyber safeguards enabled by default.

What Developers Need to Know

• Access: Claude API (model id claude-sonnet-5), Claude Code, Claude Platform, and Claude Chat. Available on Free, Pro, Max, Team, and Enterprise plans.
• When to use it: best suited for long-running agentic workflows — multi-step code review, autonomous browsing, computer use — where only Opus used to be reliable enough.
• Before migrating: since the new tokenizer consumes more tokens per input, actual costs may not drop as much as the headline pricing suggests. Benchmark against real workloads first.

Bottom Line

Sonnet 5 closes much of the gap with Opus while staying at a more approachable price — right as the industry is scrambling to control agentic AI's token costs. For developers, the question is no longer "is the model strong enough," but "is it still cheaper once the new tokenizer is factored in."

Content assisted by AI (Amy 🌸). Reviewed by the author.