ByClaude Sonnet 5 pricing is now permanent at $2 per million input tokens and $10 per million output tokens. Anthropic confirmed on August 10, 2026 that it will not go ahead with a planned increase to $3/$15 that was originally scheduled for September 1, 2026, so the launch rate that developers have used since June is now the standard, ongoing price.
That decision matters beyond the headline number. It changes how you should budget for agent workloads, what model tier makes sense for production apps, and how Sonnet 5 stacks up against Opus 5 and Haiku 4.5. This guide walks through the official numbers, what you actually get for that price, how caching and batching cut real costs further, and when Sonnet 5 is the right call over a more expensive or cheaper alternative.
When Did Anthropic Make Claude Sonnet 5 Pricing Permanent?
Claude Sonnet 5 launched in June 2026 at an introductory rate of $2 per million input tokens and $10 per million output tokens, with Anthropic stating at the time that the rate would hold through August 31, 2026 before stepping up to $3/$15. On August 10, 2026, Anthropic announced it was canceling that scheduled increase entirely, making the $2/$10 rate the model’s permanent, standard price rather than a temporary launch promotion.
The official Claude pricing documentation now lists $2/$10 as Sonnet 5’s standard rate with no expiration date attached, confirming the change directly from the source developers actually build against.
Why Did Anthropic Cancel the Planned Price Increase?
Anthropic has not published a single stated reason, but the timing lines up with two visible pressures in the market. First, competing labs shipped several new, aggressively priced models in the same window — Alibaba’s Qwen3.8 Max and Z.AI’s GLM-5.2 Turbo both launched in August 2026 — putting pressure on every provider’s mid-tier pricing. Second, Sonnet-tier models are the default choice for high-volume agent and coding workloads, where small per-token differences compound fast across millions of daily requests, so holding the line on price protects that segment of usage.
Whatever the exact motivation, the practical effect is the same for anyone building on the API: the cost model you designed around in June is now the cost model you can plan around long-term, without needing to budget for a September rate hike that will no longer happen.
What Changed Since Claude Sonnet 5 Launched in June 2026?
At launch, Sonnet 5 shipped with an explicit two-tier pricing plan: $2/$10 as an introductory rate through the end of August, then $3/$15 afterward. That structure is common for new model releases — it lets a lab gather real usage data and gauge demand before locking in a permanent rate, while giving early adopters a clear incentive to test the model quickly.
Between June and August 2026, Sonnet 5 became the default choice for a large share of coding-agent and production-agent traffic on the Claude API, based on its position as the mid-tier model built specifically for that use case. Canceling the planned increase rather than letting it take effect and adjusting later suggests Anthropic viewed the $2/$10 rate as core to that adoption, not just a promotional hook. See our Claude AI 2026 advancements roundup for how Sonnet 5 fits into Anthropic’s broader model lineup this year.
How Does Claude Sonnet 5 Pricing Compare to Opus 5 and Haiku 4.5?
Anthropic currently sells three active model tiers, each priced for a different balance of capability and cost. Here is how Claude Sonnet 5 pricing sits between them, using the standard API rates per million tokens.
| Model | Input (per MTok) | Output (per MTok) | Batch API (in/out) | Best for |
|---|---|---|---|---|
| Claude Opus 5 | $5 | $25 | $2.50 / $12.50 | The hardest reasoning, coding, and agent tasks |
| Claude Sonnet 5 | $2 | $10 | $1 / $5 | Most production apps and agent workloads |
| Claude Haiku 4.5 | $1 | $5 | $0.50 / $2.50 | High-volume, latency-sensitive, simple tasks |
Sonnet 5 costs 2.5x less than Opus 5 on both input and output tokens, while Haiku 4.5 costs half of Sonnet 5. That spread is intentional: Anthropic positions Sonnet 5 as the tier most teams should default to for production work, reserving Opus 5 for tasks that genuinely need its extra reasoning depth and Haiku 4.5 for simple, high-volume calls where speed and cost matter more than raw capability. Our Claude Opus 5 guide covers what that extra reasoning depth actually looks like in practice.
What Do You Actually Get for $2/$10 Per Million Tokens?
Claude Sonnet 5 pricing covers a model built for agentic and coding work at production scale: long-running multi-step tasks, tool use, and codebase-level reasoning, at a cost low enough to run continuously rather than sparingly. It includes the full 1M-token context window at standard per-token pricing, so a 900,000-token request costs the same per token as a 9,000-token request — there is no long-context surcharge tier to plan around.
One detail worth budgeting for: Sonnet 5 uses an updated tokenizer that produces more tokens for the same input text than older Claude models did, roughly 1.0–1.35x depending on content type. That means a straight per-token price comparison against an older model can understate your real cost difference — test with your actual prompts rather than assuming token counts transfer directly from a previous model.
How Does Prompt Caching Change the Real Cost of Claude Sonnet 5?
Prompt caching is where Claude Sonnet 5 pricing gets meaningfully cheaper in practice. A 5-minute cache write costs 1.25x the standard input rate, and a 1-hour cache write costs 2x, but a cache read costs only 0.1x the standard input price — a 90% discount on any content you reuse across requests.
For an agent that repeatedly sends the same system prompt, tool definitions, or a large document as context, caching turns Sonnet 5’s already-low $2 input rate into an effective $0.20 per million tokens on every cached hit after the first. Any workload with a stable, reused context block — a coding agent working in the same repo, a support bot with a fixed knowledge base — should have caching enabled by default.
When Should You Use Claude Sonnet 5 Instead of Opus 5?
Default to Sonnet 5 for the majority of production workloads: coding assistants, customer-facing agents, document analysis, and most multi-step tool-use pipelines. Its price-to-capability ratio is built specifically for running continuously in these scenarios rather than being reserved for occasional, expensive queries.
- Choose Sonnet 5 for day-to-day coding agents, chat products, and workflows where cost scales with volume.
- Choose Opus 5 for the hardest reasoning steps — complex debugging, architectural decisions, or tasks where a wrong answer is expensive to unwind — and consider routing only those specific steps to Opus while the rest of the pipeline stays on Sonnet.
- Choose Haiku 4.5 for classification, extraction, and simple routing calls where latency and cost matter more than depth.
A common production pattern is a mixed pipeline: Haiku 4.5 for cheap triage, Sonnet 5 for the main workload, and Opus 5 called selectively for the subset of requests that are flagged as high-complexity. That structure captures most of Opus 5’s quality where it matters while keeping the bulk of your token spend at Sonnet 5’s lower rate.
How Do You Estimate Your Own Monthly Cost With Claude Sonnet 5?
Start with three numbers: average input tokens per request, average output tokens per request, and requests per month. Multiply input tokens by $2 per million, output tokens by $10 per million, then multiply by your monthly volume. If a meaningful share of your input is repeated context, subtract the caching discount for that portion before totaling.
A worked example: a coding agent that averages 8,000 input tokens and 2,000 output tokens per task, running 50,000 tasks a month, with half of the input coming from a cached repository context read on every call.
| Line item | Calculation | Monthly cost |
|---|---|---|
| Uncached input (4,000 tok x 50,000) | 200M tokens x $2 / MTok | $400 |
| Cached input reads (4,000 tok x 50,000) | 200M tokens x $0.20 / MTok | $40 |
| Output tokens (2,000 tok x 50,000) | 100M tokens x $10 / MTok | $1,000 |
| Total | $1,440 |
Without caching, that same workload would cost roughly $1,800 a month — caching alone saves about 20% here, and the savings grow with a higher proportion of reused context. Run this same math with your own token counts from the Claude Console’s usage dashboard rather than estimates, since actual tokenized length can differ from a rough word-count guess.
The same formula scales linearly, which makes it easy to model growth. Doubling monthly task volume to 100,000 roughly doubles the total to about $2,880 with caching in place, assuming the same per-task token averages and cache-hit ratio hold. If your cache-hit ratio improves as your context becomes more stable over time — a common pattern once an agent’s system prompt and tool definitions stop changing — the effective cost per task drops further even as raw volume grows.
Key numbers worth keeping on hand when you plan a budget around Claude Sonnet 5 pricing:
- Standard API: $2 / MTok input, $10 / MTok output — permanent as of August 10, 2026.
- Batch API: $1 / MTok input, $5 / MTok output for asynchronous jobs.
- Cache reads: $0.20 / MTok, a 90% discount versus standard input.
- Full 1M-token context window at standard per-token pricing, no long-context surcharge.
- US-only inference routing: 1.1x multiplier across all token categories.
If you are building the agent itself rather than just calling the API, our guide to AI coding assistants in 2026 compares how Sonnet-tier models perform across real coding benchmarks, which is useful context alongside the raw pricing numbers here.
Does Data Residency Change Claude Sonnet 5 Pricing?
Yes, if you pin inference to a specific region. Global routing — the default for the Claude API — uses the standard $2/$10 rate with no surcharge. Requesting US-only inference through the inference_geo parameter applies a 1.1x multiplier across every pricing category: input tokens, output tokens, cache writes, and cache reads. The same 1.1x premium applies to the equivalent US Data Zone deployment option on Claude in Microsoft Foundry.
For most teams outside regulated industries, global routing is fine and keeps costs at the standard rate. Healthcare, finance, and government-adjacent workloads that require guaranteed in-country processing should factor the 10% premium into their budget rather than being surprised by it after their first invoice.
Can You Try Claude Sonnet 5 Before Committing to a Budget?
New API accounts receive a small amount of free credit to test Claude Sonnet 5 and other models before any billing begins, which is enough to run the kind of cost estimate described above against your own real prompts rather than guessing. Anthropic also offers extended evaluation arrangements for enterprise teams that want to pilot Sonnet 5 across a larger workload before signing a production contract — worth requesting through sales rather than assuming only the small self-serve credit is available.
Where Can You Use Claude Sonnet 5 Beyond the API?
Claude Sonnet 5 is available directly through the Claude API, through Claude Platform on AWS Marketplace (billed in Claude Consumption Units rather than a per-token line item), on Amazon Bedrock, and on Google Cloud Vertex AI, each with its own billing wrapper around the same underlying model. If your team is already inside AWS or Google Cloud billing, those platforms can be simpler for procurement even though the effective per-token math traces back to the same standard rate.
For WordPress specifically, connecting Claude to your site’s admin workflow is a practical way to put Sonnet 5’s lower cost to work on content and site-management tasks. Our walkthrough on using the Claude connector in WordPress covers the setup end to end.
What Additional Tool Costs Should You Budget For?
Tool use itself adds a small, fixed token overhead on top of your prompt and response tokens. For Claude Sonnet 5, enabling tools adds roughly 354 tokens to the system prompt with automatic tool choice, or about 474 tokens when you force a specific tool call — both billed as ordinary input tokens at the standard $2 rate, not as a separate line item.
Heavier tool sets cost more overhead. Declaring the full computer-use toolset adds about 4,590 input tokens per request on Sonnet 5, and the full browser-use toolset adds about 6,670 tokens, figures that are slightly higher on Sonnet 5 than on Opus 5 due to its tokenizer. Screenshots returned by either tool are billed as image input on top of that. None of this changes the per-token rate itself, but it is easy to under-budget an agent that leans heavily on computer or browser use if you only price out its text prompts.
Frequently Asked Questions About Claude Sonnet 5 Pricing
Is Claude Sonnet 5 pricing the same on every platform?
The underlying per-token rate is consistent, but the billing wrapper differs. The first-party Claude API bills per million tokens directly. Claude Platform on AWS and Azure bill in Claude Consumption Units at $0.01 each, converted from the same standard token rate. Bedrock and Vertex AI have their own marketplace pricing pages that mirror the standard rate in most cases.
Does Claude Sonnet 5 pricing include the 1M-token context window?
Yes. Claude Sonnet 5 includes the full 1M-token context window at standard per-token pricing, with no separate long-context rate tier. A large request is billed at exactly the same per-token rate as a small one.
Is Claude Sonnet 5 cheaper to run than Claude Sonnet 4.5?
Per token, yes — Sonnet 5 is priced at $2/$10 versus Sonnet 4.5’s $3/$15. However, Sonnet 5’s updated tokenizer produces more tokens for the same input text, so the real-world cost gap is smaller than the sticker price suggests. Test against your actual prompts before assuming the full discount applies.
Will Anthropic raise Claude Sonnet 5 pricing again later?
Anthropic has stated the $2/$10 rate is now standard pricing, not a temporary promotion, and has not announced any future increase. As with any API pricing, it is worth checking the official pricing page periodically rather than assuming it is permanent forever.
How much does Claude Sonnet 5’s web search tool cost on top of tokens?
Web search is billed separately at $10 per 1,000 searches, plus the standard token cost for whatever search-generated content gets added to the conversation. The code execution and web fetch tools carry no additional charge when used alongside web search, beyond standard token costs.
Does the Batch API reduce Claude Sonnet 5 pricing further?
Yes. The Batch API applies a 50% discount on both input and output tokens for non-time-sensitive workloads, bringing Sonnet 5 down to $1 per million input tokens and $5 per million output tokens. It is worth using for any job that does not need a synchronous response, such as bulk content generation or offline analysis.
Are there volume discounts available beyond standard Claude Sonnet 5 pricing?
Standard published rates apply to most accounts, but Anthropic negotiates volume discounts case by case for high-usage customers, alongside separate academic and research discount programs. If your monthly spend is significant, contacting the Anthropic sales team directly is worth doing before assuming the listed per-token rate is your final cost, since enterprise agreements are not published publicly and vary by commitment size.
Subscribe for Newsletter
