Claude Sonnet 5: What's New, Pricing, and Why It Matters
The mid-tier model just stopped being the compromise.
Claude Sonnet 5 shipped on June 30, 2026, and it rearranges the usual mental model of Anthropic's lineup. Sonnet used to be the sensible middle child: cheaper than Opus, smarter than Haiku, picked mostly by people watching their bill. Sonnet 5 changes the pitch — Anthropic says its performance is close to Opus 4.8, it ships a 1 million token context window by default, and it launched at an introductory price of $2 per million input tokens. The middle child suddenly has main-character energy.
It is also now the default model for Free and Pro users on claude.ai, which means Sonnet 5 is the Claude most of the world will actually talk to in 2026. Here is everything that changed, including the tokenizer detail that will quietly affect your costs.
- Released June 30, 2026; now the default Claude for Free and Pro plans.
- 1M-token context window is the default and the maximum, with 128K max output.
- Intro pricing: $2/$10 per million tokens through August 31, 2026 — then $3/$15.
- New tokenizer produces roughly 30% more tokens from the same text; intro pricing offsets it.
- Adaptive thinking is on by default — the model decides how hard to think per request.
What's new in Claude Sonnet 5?
Three changes carry most of the weight.
1. A million tokens of context, standard
Previous Sonnet generations treated long context as an option or a beta. Claude Sonnet 5 makes 1M tokens the default and the ceiling, with up to 128K tokens of output. In practice that is an entire codebase, a legal discovery folder, or a season of podcast transcripts in a single conversation. Combined with file attachments in CoreAI's chat, "just upload all of it" is now a legitimate workflow rather than a way to get scolded by a token limit.
2. Adaptive thinking by default
Sonnet 5 decides for itself how much reasoning a request needs — quick answers stay quick, hard problems get genuine deliberation. Anthropic committed to this fully: manual extended-thinking flags now return an error on this model. Fewer dials to configure, and in most hands, better results than manually guessing a thinking budget.
3. Agentic reliability
The less flashy but arguably most important upgrade: better instruction following, tool selection, and error correction across long multi-step tasks. Sonnet 5 is built for the 40-minute autonomous workflow, not just the 40-second answer — it recovers from its own mistakes instead of compounding them, which is precisely the trait that separates a demo from a dependable assistant.
How much does Claude Sonnet 5 cost?
Pricing comes with a twist worth understanding before you compare providers:
| Period | Input (per 1M tokens) | Output (per 1M tokens) |
|---|---|---|
| Intro — through Aug 31, 2026 | $2 | $10 |
| From Sep 1, 2026 | $3 | $15 |
| Claude Sonnet 4.6 (reference) | $3 | $15 |
The twist: Sonnet 5 uses a new tokenizer that produces about 30% more tokens from the same input text than Sonnet 4.6 did. The intro pricing is calibrated so the transition is roughly cost-neutral — you pay less per token but count more tokens. After August 31, identical workloads will cost moderately more than they did on 4.6. If you run high-volume pipelines, that is worth budgeting for now rather than discovering on an invoice in September.
Is Claude Sonnet 5 better than Sonnet 4.6?
For nearly everything, yes — this is a generational jump, not a patch. Writing quality keeps the trademark Sonnet voice that made Sonnet 4.6 our pick for writing last generation, while reasoning, coding, and long-context recall move decisively upward. The places to stay skeptical:
- Cost-sensitive pipelines: after the intro window ends, the tokenizer change means real-world costs rise versus 4.6. Cheap-tier work may belong on a budget model anyway.
- Manual thinking control: if your workflow depended on forcing extended thinking on or off, adaptive-only behavior is a change you will feel.
- Peak difficulty: "close to Opus 4.8" is Anthropic's phrasing, and it is honest — on the hardest reasoning problems, Opus 4.8 and the new frontier tier still win. See our Sonnet vs Opus enterprise guide for how to think about that split.
The right way to settle it for your own work: run the same prompt through Sonnet 5, Sonnet 4.6, and a rival on CoreAI's side-by-side Compare. Model marketing is a genre; your own prompt is data.
Where does Sonnet 5 fit in a crowded July?
Anthropic did not launch into a quiet news cycle: xAI shipped Grok 4.5 on July 8 and OpenAI answered with GPT-5.6 a day later. Sonnet 5's differentiators in that field are the 1M context window as a default (not an upsell) and the agentic reliability work. Its price sits between GPT-5.6 Luna and Terra while — per Anthropic's own detailed release notes at anthropic.com — punching near Opus-class weight. For anyone whose work involves long documents, that combination is currently unique at this price.
It is also, notably, the free default on claude.ai — and available on CoreAI alongside every rival mentioned above, which is a better deal if you want the whole field in one place. Browse the full lineup in the model library.
Who should switch to Sonnet 5 today?
If you are on Sonnet 4.6, switch now and enjoy the intro pricing — the upgrade is free lunch until September, and the long-context ceiling alone justifies the move for research, legal, and documentation work. If you are on a rival stack, the calculus depends on your workload: writers and analysts gain the most from the 1M window and Claude’s prose quality, while pure high-volume pipelines may still be cheaper on budget tiers like GPT-5.6 Luna or DeepSeek V4 Flash. And if you run agents, the error-correction improvements are the quiet headline — fewer stuck loops means fewer babysitting sessions. Teams already juggling several providers can skip the migration debate entirely: on CoreAI you can point the same conversation at Sonnet 5 one message and a rival the next, and let results decide.
Frequently Asked Questions
When was Claude Sonnet 5 released?
June 30, 2026. It immediately became the default model for Free and Pro users on claude.ai, and is available to Max, Team, and Enterprise plans, as well as through the API and on CoreAI.
What is Claude Sonnet 5's context window?
1 million tokens — and unlike previous generations, that is the default configuration, not a beta or an add-on. Maximum output is 128K tokens per response.
How much does Claude Sonnet 5 cost?
Introductory API pricing is $2 per million input tokens and $10 per million output through August 31, 2026, rising to $3/$15 afterward. Note the new tokenizer counts roughly 30% more tokens for the same text compared to Sonnet 4.6.
Is Claude Sonnet 5 as good as Opus?
Anthropic describes its performance as close to Claude Opus 4.8 at a lower price. For everyday professional work the gap is hard to notice; for the very hardest reasoning tasks, Opus-class models retain the edge.
Can I use Claude Sonnet 5 without a Claude subscription?
Yes — Claude Sonnet 5 is live on CoreAI for web, iOS, and Android, under one subscription that also covers GPT-5.6, Gemini, Grok 4.5, and 300+ other models.
Try Claude Sonnet 5 on CoreAI today
One subscription, 300+ models — Sonnet 5, GPT-5.6, Grok 4.5 and every other frontier model, side by side.