Claude Sonnet 5 arrives — near-Opus 4.8 quality at ~40% the sticker price, now the default for Free and Pro (mind the tokenizer)

What launched

Per Anthropic’s own announcement and coverage from DataCamp, Modem Guides, and Advisori:

• What it is: Claude Sonnet 5, Anthropic’s new mid-tier model — “built to be the most agentic Sonnet model yet,” narrowing the gap with flagship Opus 4.8.
• Where it’s live: the default for Free and Pro on claude.ai; available to Max, Team, Enterprise; and in Claude Code, the Claude API (claude-sonnet-5), Cursor, and GitHub Copilot.
• Pricing: $2 input / $10 output per million tokens through August 31, then $3/$15. (Opus 4.8: $5/$25.)
• Context: a 1M-token window.
• Capability: “matches Opus 4.8’s capability levels” in some cases and “covers a much wider range of cost-performance options,” per Anthropic. Reported third-party benchmarks put it modestly behind Opus 4.8 on agentic coding (~63% vs ~69% SWE-bench Pro) but ahead on some agentic tasks like Terminal-Bench.
• The tokenizer change: a new tokenizer maps the same text to ~1.0-1.35× more tokens; Anthropic set intro pricing so the transition is “roughly cost-neutral.”

Why this matters

Three reads.

1. This is the model most people will actually use — and it just got much more capable. Opus grabs headlines and Fable 5 gets government attention, but Sonnet is the workhorse: the default tier for the vast majority of Claude usage. Making Sonnet 5 the default for Free and Pro means Anthropic just upgraded the everyday Claude experience for its largest user base, with agentic gains that matter for the real work people do — coding, tool use, multi-step tasks. For most readers, “Claude got better this week” is literally true, at no action required.

2. The tokenizer caveat is the story the press releases bury — and it’s the useful part. “Near-Opus quality at 40% of the price” is the headline everyone will run. The honest version is more nuanced: because Sonnet 5 counts more tokens per unit of text, and you pay per token, a lower rate paired with more tokens partly cancels. Anthropic itself says the switch is “roughly cost-neutral” versus Sonnet 4.6 — a candid admission that this is primarily a capability upgrade, not a cost cut. Anyone budgeting API spend should measure tokens on their own workloads rather than trusting the rate card. That’s the read that beats the breathless coverage.

3. It’s Anthropic executing the “efficiency over tokenmaxxing” shift. The market has been moving from “use the biggest model” to “get the most capability per dollar” — a trend even CNBC has flagged. Sonnet 5 is Anthropic’s answer: a model tuned to sit just below the flagship on capability while opening a much wider cost-performance range. Combined with the just-restored Fable 5 at the top and Opus 4.8 in the middle, Anthropic now has a cleaner tiered lineup — flagship, workhorse, and frontier — mirroring OpenAI’s new Sol/Terra/Luna split. Tiering is now the industry default.

What it means for you

If you’re on Claude Free or Pro: you already have Sonnet 5 — it’s the default. Expect noticeably better agentic behavior (tool use, multi-step tasks, coding) than Sonnet 4.6, with no change to your subscription. This is the rare “your tool improved overnight” moment.

If you use Claude Code, Cursor, or Copilot: Sonnet 5 is available as a model option and is a strong pick for everyday coding — close to Opus 4.8 on many tasks at a lower rate. For the hardest long-horizon work, Opus 4.8 or the restored Fable 5 still lead; Sonnet 5 is the efficient default for the other 80%.

If you pay by API token: run the numbers on your workloads. The $2/$10 intro rate is attractive, but with the tokenizer counting more tokens and standard pricing rising to $3/$15 after August 31, your real per-task cost may be close to what you paid on Sonnet 4.6. Budget on measured token usage, not the headline rate.

If you’re choosing between Claude, ChatGPT, and Gemini: Sonnet 5 strengthens Claude’s mid-tier value case. See the Claude vs ChatGPT and best AI chatbots guides for where each wins.

The honest caveats

The “40% cheaper than Opus” framing is misleading in isolation. Sonnet 5 and Opus 4.8 are different tiers, not the same model at different prices — comparing rate cards without comparing capability (and token counts) overstates the savings. The right comparison for cost is Sonnet 5 vs Sonnet 4.6, and there Anthropic says it’s roughly cost-neutral.

Third-party benchmarks are early and mostly from aggregators. Anthropic’s own page emphasizes agentic-search and computer-use gains without a full published scorecard; the specific SWE-bench/Terminal-Bench numbers circulating come from secondary sources and should be treated as indicative until independent evals settle.

Intro pricing expires. The attractive $2/$10 rate runs only through August 31; standard $3/$15 is the real long-run price. Factor that into any cost planning.

“Most agentic Sonnet yet” is a positioning claim. It’s credible given the benchmark direction, but “most agentic” is Anthropic’s framing. Real-world agentic reliability is what matters, and that’s proven in use, not launch copy.

What it changes for Pick Right readers

The practical takeaway is genuinely good: the default Claude just got better for free, and for coding and agentic work Sonnet 5 is now the sensible everyday pick — with Opus 4.8 and the restored Fable 5 reserved for the hardest tasks. Just don’t read the rate card as a discount: it’s a capability upgrade priced to be roughly cost-neutral, with the tokenizer doing quiet work behind the numbers. We’ve updated the Claude review to reflect Sonnet 5 as the new default tier.

For more, see the Claude review, the Claude Code review, the Opus 4.8 launch, the Fable 5 restoration, the OpenAI GPT-5.6 Sol/Terra/Luna preview (the parallel tiering move), the Claude vs ChatGPT comparison, and the best AI chatbots guide.

Questions or corrections? Email Pick Right. Want the full list? See all news.