Claude Sonnet 5: Anthropic's workhorse at a promo price of 2/10 dollars
Anthropic released Claude Sonnet 5 on July 1, 2026 — the new version of the balanced mid-tier line that sits between fast Haiku and expensive Opus. The vendor positions the model as an everyday workhorse and markets it as delivering performance close to the Opus class at a much lower price. At launch, Anthropic is charging a promotional price of 2 US dollars per million input tokens and 10 US dollars per million output tokens — explicitly labeled as an introductory rate. On the same day, Anthropic also made Claude Fable 5 globally available again after the US export-control pause.
What Anthropic states at launch
- Released July 1, 2026 — Sonnet 5 replaces the Sonnet 4.x line as the current mid-tier.
- Intro price (promotional): 2 US dollars per million input tokens, 10 US dollars per million output tokens.
- Positioning: per the vendor, near-Opus performance, placed between Haiku (fast/cheap) and Opus (top tier).
- Target use: agent and coding workflows that burn through many tokens.
- In parallel: Claude Fable 5 available worldwide again after the export-control pause.
Where things stood
The Sonnet line has always been Anthropic’s mid-tier: faster and cheaper than Opus, more capable than Haiku. In practice, Sonnet was the default model for the bulk of tasks where Opus was overkill and Haiku too weak. With Sonnet 4.5 (2025) and Sonnet 4.6 (2026), Anthropic had updated the line at short intervals without changing the basic price structure much.
Cost was always the argument for Sonnet: anyone running agents or coding assistants quickly burns tens of thousands of tokens per task. At Opus prices of 5/25 US dollars per million tokens, that adds up fast — Sonnet was the answer to keeping usable quality affordable.
What is new
1. A new promo price, clearly labeled as such. At 2 US dollars input and 10 US dollars output per million tokens, Sonnet 5 launches well below Opus prices. Important for planning: Anthropic itself calls this an introductory price. If you build a budget on it, factor in that it can rise after the promo phase — the price increase is already on the horizon and is covered separately.
2. Near-Opus performance — as a vendor claim. Anthropic markets Sonnet 5 as reaching quality close to the Opus class. That is a vendor statement, not an independent benchmark. The takeaway: plausible as positioning, but verify on your own tasks before switching, rather than trusting the marketing.
3. Sonnet 5 clearly targets token-heavy workflows. The combination of a mid-tier model and a low intro price is aimed exactly at cases where volume matters: agents that take many steps, coding assistants that read entire repositories, batch processing of large text volumes. Here the price per token directly decides the economics.
Reading
The real lever with Sonnet 5 is not performance but price — and that is exactly where a closer look pays off. An introductory price is a marketing instrument: it lowers the barrier to switching and ties users to the model before the regular price kicks in. For agencies and teams that need stable cost calculations over months, a promotional price is therefore not a reliable planning figure.
Recent Anthropic history supports this caution: over the past months the company has repeatedly adjusted price structure, plan limits, and billing (see our news on extra-usage billing and volume-tier discounts). A low launch price fits this pattern — it is part of a pricing strategy, not a permanent promise. Tellingly, a price increase for late August is already on the table.
For model selection, the sober rule stands: Sonnet 5 is a strong candidate for the mid-tier task range, but the decision should be based on the regular price, not the promo. Switching now wins in the short term — but build the move so a later price jump does not topple your whole calculation.
What you can do now
If you want to test Sonnet 5: use the promo phase to measure on your real tasks whether the near-Opus claim holds for your cases. Compare output quality and token consumption against your current model — not just the price.
If you plan budgets: calculate with the expected regular price, not the intro price. A cost model that only works at 2/10 dollars is fragile.
If you want to avoid model lock-in: build your workflows so a model switch stays possible without a major rebuild. Especially with price promotions, the ability to switch back is the real protection.