Quick Answer
Anthropic made Claude Sonnet 5’s introductory API rate of $2 per million input tokens and $10 per million output tokens permanent on August 10, 2026, canceling a scheduled increase to $3/$15 that was set to take effect September 1. Subscription plan prices are unchanged. API users should verify their budgets against Anthropic’s official pricing page, because a new tokenizer can increase billable token counts by up to 35 percent.
Key Takeaways
- The $2/$10 per million input/output token rate for Claude Sonnet 5 is now Anthropic’s standard price, not a limited-time offer.
- The previously scheduled September 1, 2026, increase to $3/$15 per million tokens will not occur, according to Anthropic’s official platform documentation.
- Claude Sonnet 5 is the default model for Free and Pro subscribers on claude.ai and requires no action to access at the new permanent rate.
- A new tokenizer in Sonnet 5 can generate up to 35 percent more billable tokens from the same input text compared to Sonnet 4.6, so the effective cost per task may be higher than the per-token rate suggests.
- Competing models including GPT-5.6 Luna ($1/$6) and GPT-5.6 Terra ($2.50/$15) sit near Sonnet 5 on price; the most accurate cost comparison is cost per completed task, not per-token sticker price.
What exactly did Anthropic change about Claude Sonnet 5 pricing?
The $2/$10 per million input/output token pricing for Claude Sonnet 5, announced at launch as introductory pricing through August 31, 2026, is now the standard price. Anthropic confirmed the change in an official post on August 10, 2026, stating directly that the introductory pricing is now permanent, with the $2 per million input and $10 per million output rate remaining unchanged.
The previously scheduled increase to $3/$15 per million input/output tokens on September 1, 2026, will not occur. That planned step-up represented a 50 percent increase over the introductory rate on every token, so the cancellation is a meaningful change for any team that had already budgeted for the higher September rate. Developers who adjusted cost models to absorb that increase can now revert those projections, though they should confirm the current rate directly on Anthropic’s official platform pricing page before finalizing any budget.
How does this affect Free, Pro, and Max subscribers?
The pricing change applies to the Claude API, which is the pay-per-token interface developers use to build applications. Subscription plans for claude.ai are separate and not affected by the per-token rate announcement. Sonnet 5 is the default model for Free and Pro plans, and is also available on Max, Team, Enterprise, Claude Code, and the Claude API.
Claude Pro is $20 per month, or $17 per month billed annually, and runs Sonnet 5 by default with access to more Claude models, more usage than Free, Research mode, unlimited Projects, voice mode, and Microsoft 365 integration. Claude Max costs $100 per month for the Max 5x tier or $200 per month for the Max 20x tier, with Max 5x providing 25 times the capacity of the Free plan and Max 20x providing 100 times the Free plan capacity. For subscribers on any of these plans, no action is required. Sonnet 5 was already the default model, and subscription pricing has not changed.
What is Claude Sonnet 5 and why does it matter?
Claude Sonnet 5 launched June 30, 2026, replacing Sonnet 4.6 as Anthropic’s mid-tier agentic model. Anthropic describes it as its most agentic Sonnet yet, noting that it makes plans, uses tools like browsers and terminals, and runs autonomously at a level that previously required larger and more expensive models. That capability description is Anthropic’s own characterization; independent benchmark evaluations of those claims remain limited at this stage.
Claude Sonnet 5 ships with a 1M-token context window and, according to Anthropic, lands close to Opus 4.8 quality for considerably less money. The model offers a 1M-token context window and 128K maximum output, matching the headline numbers of Opus 4.8 and Fable 5, and it is the first time the mid-tier has had frontier-scale context as a native feature without any asterisks. Because AI models are now being deployed in complex, autonomous workflows, the combination of near-flagship capability and a now-permanent lower price point has direct implications for teams running high-volume agentic tasks.
What is the tokenizer issue and how does it affect your actual bill?
The permanent $2/$10 rate is the most prominent number in this announcement, but it is not the only number that determines what API users actually pay. Sonnet 5 uses a new tokenizer that produces roughly 1.0 to 1.35 times more tokens per input than Sonnet 4.6, so the effective cost per unit of text runs higher than the sticker rate. This matters because per-token pricing multiplies token count by the per-token rate, and a larger token count raises the bill even when the rate itself is lower.
Sonnet 5’s updated tokenizer may encode the same prompt as 1.0 to 1.35 times more tokens than Sonnet 4.6, meaning that for a typical 1,000-token Sonnet 4.6 prompt, developers should plan for up to 1,350 tokens on Sonnet 5. The tokenizer effect is typically most pronounced on code, structured data, and non-English text. The practical response for developers migrating workloads from Sonnet 4.6 is to measure token counts on real production prompts before finalizing cost projections, not to rely on the headline per-token rate alone. Two cost-reduction mechanisms are also available: cache hits cost 10 percent of the standard input rate, and the Batch API cuts input and output prices by 50 percent.
How does Claude Sonnet 5’s permanent price compare to competing models?
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Availability (August 2026) |
|---|---|---|---|
| Claude Sonnet 5 | $2.00 | $10.00 | Generally available |
| Claude Opus 5 | $5.00 | $25.00 | Generally available |
| Claude Haiku 4.5 | $1.00 | $5.00 | Generally available |
| GPT-5.6 Sol (OpenAI) | $5.00 | $30.00 | Limited preview |
| GPT-5.6 Terra (OpenAI) | $2.50 | $15.00 | Limited preview |
| GPT-5.6 Luna (OpenAI) | $1.00 | $6.00 | Limited preview |
All pricing figures sourced from vendor-published rates as of August 2026. Verify current rates on each provider’s official pricing page before committing to a budget. GPT-5.6 figures are sourced from Emergent.sh’s July 2026 comparison; Claude figures sourced from Anthropic’s platform documentation.
GPT-5.6 is a model family, not a single model, comprising Sol at $5/$30 per million tokens, Terra at $2.50/$15, and Luna at $1/$6. At the per-token level, GPT-5.6 Luna undercuts Sonnet 5 on input cost, while Terra is slightly more expensive on input but identical on output at $15 per million. That said, on published coding benchmarks the gap between Sonnet 5 and GPT-5.6 Terra is narrow where it counts: the two models score essentially the same on SWE-bench Pro (63.2 percent versus 63.4 percent), though both figures are vendor-reported. Independent evaluations of those scores should be treated with caution, as OpenAI published an audit in July 2026 finding that approximately 30 percent of SWE-bench Pro tasks contain flawed test criteria, incomplete descriptions, or misleading task statements. The more relevant comparison for most teams is cost per completed task on their own workloads, not per-token sticker price or a single benchmark number.
What does the permanent rate mean for developers running agents in production?
For developers building on the Claude API, the permanent rate removes a planning uncertainty that had complicated multi-month cost projections. Teams that had already modeled the September 1 step-up to $3/$15 can now restore the $2/$10 baseline as a stable, long-term assumption. The Batch API halves both rates, and prompt caching can cut cached input by up to 90 percent. Stacking those discounts on the now-permanent base rate gives high-volume teams a meaningful cost floor, provided their workloads can tolerate batch processing delays and involve repeated context across calls.
The practical limit for agentic workloads is token volume, not token rate. Claude Sonnet 5 is Anthropic’s mid-tier model built for production coding, multi-step agents, and long-running software tasks, and it pairs a 1 million-token context window with a 128,000-token maximum output. An agent that repeatedly reads large codebases or documents accumulates tokens rapidly across turns, which means the tokenizer’s 1.0 to 1.35 multiplier compounds across those turns. Developers should benchmark token usage per agent run on representative production tasks before treating the $2/$10 headline as a definitive budget figure. For those building AI-powered developer tools, it is also worth noting that AI coding agents carry their own security considerations independent of the pricing question.
Should Free and Pro subscribers do anything differently after this announcement?
Free and Pro subscribers on claude.ai do not need to take any action as a result of this announcement. Claude Sonnet 5 has been the default model on both the Free and Pro plans since June 30, 2026, and it includes features such as Projects, file uploads, and Memory. The permanent rate affects only API billing, not the flat-fee subscription structure. Subscription costs remain $0 for Free, $20 per month for Pro (or $17 per month billed annually), and $100 or $200 per month for Max 5x and Max 20x respectively, as Anthropic confirmed in its current pricing documentation.
For users evaluating whether to move from the Free plan to a paid subscription, the more relevant question is usage volume, not the token rate change. The free plan carries usage limits and feature restrictions that are worth understanding before hitting them mid-task. Users who consistently reach the Free plan’s message ceiling should consider Claude Pro at $20 per month, while those who rely on Claude for several hours of intensive daily work may find the Max 5x tier at $100 per month better suited to avoid mid-session interruptions. Confirm the current plan details directly on Anthropic’s official pricing page at claude.ai/pricing before upgrading, as usage limits and included features have shifted several times during 2026. The broader AI software cost conversation also extends to what these models run on, which is worth understanding for teams planning hardware investments alongside API budgets.
FAQ
Is the Claude Sonnet 5 price change permanent or could Anthropic raise it again?
Anthropic stated in its August 10, 2026, announcement that the $2/$10 rate is now the standard price, removing the previous August 31 expiration. Anthropic can adjust its pricing at any time, so the most reliable approach is to monitor the official platform pricing page for any future changes rather than treating any rate as guaranteed indefinitely.
Does the permanent $2/$10 rate apply to my claude.ai subscription, or only to the API?
The $2/$10 per million token rate applies only to the Claude API, which developers use to build applications with pay-per-token billing. Claude Free, Pro, Max, Team, and Enterprise subscriptions are flat-fee plans priced separately, and those subscription prices were not changed by this announcement.
Will the Claude Sonnet 5 tokenizer increase my API bill even at the lower rate?
Yes, the new tokenizer can increase billable token counts by up to 35 percent compared to Sonnet 4.6 on the same input text, particularly for code, structured data, and non-English content. Developers migrating from Sonnet 4.6 should measure token counts on real production prompts to understand their actual effective cost before finalizing API budgets.
How does Claude Sonnet 5 compare to GPT-5.6 on price and performance?
Claude Sonnet 5 at $2/$10 per million tokens sits between GPT-5.6 Luna ($1/$6) and GPT-5.6 Terra ($2.50/$15) on input price. On published coding benchmarks, Sonnet 5 and GPT-5.6 Terra score within one percentage point of each other on SWE-bench Pro, though both scores are vendor-reported and independent evaluations remain limited. The most accurate comparison for any team is cost per completed task on their own workloads, not per-token sticker price alone.
What model do I get on the Claude Free plan, and is it affected by this pricing change?
Claude Sonnet 5 is the default model on the Claude Free plan and has been since June 30, 2026. The permanent rate announcement does not change what Free plan users receive or how usage limits work. Free plan access to Sonnet 5 is subject to Anthropic’s daily message limits, which vary based on prompt complexity, and is available at claude.ai without a payment method.