Last updated: June 30, 2026
Key Takeaways
- Claude Sonnet 5 launched June 30, 2026, and is now the default model for Free and Pro plans, priced at $2 per million input tokens and $10 per million output tokens through August 31, 2026 ($3 / $15 standard after that date).
- Benchmarks land close to Opus 4.8, for example 63.2 percent versus 69.2 percent on agentic coding, at roughly 40 percent of Opus's standard price, with a full 1 million token context window.
- Sonnet 5 is effectively the best Claude model most people can use right now, because Mythos 5 and Fable 5 remain offline under a US export-control order with no confirmed return date.
Anthropic released Claude Sonnet 5 today. It is the default model for every Free and Pro user starting now, and it is also available to Max, Team, and Enterprise users, in Claude Code, and through the Claude Platform API as claude-sonnet-5. Anthropic's own benchmarks put it close to Opus 4.8 on agentic coding, reasoning, and computer-use tasks, at a fraction of Opus's standard price.
That is the version of this story every outlet is running today. The more useful version, the one this article gets to, is what Sonnet 5's release actually means once you account for where it sits in Anthropic's current lineup. The company's two most capable models, Mythos 5 and Fable 5, have been offline under a US government export-control order since June 12, with no confirmed return date. Until that changes, Sonnet 5 is not just an upgrade. For nearly everyone reading this, it is the best Claude model you can actually use.
What's Actually New
Claude Sonnet 5 carries a standard 1 million token context window and a 128,000 token output limit, the same ceiling as Claude Sonnet 4.6. It supports the same tools and platform features as Sonnet 4.6, with one exception: Priority Tier is not available on Sonnet 5. Anthropic also raised rate limits across Chat, Cowork, Claude Code, and the Claude Platform to cover the heavier token usage of its higher effort levels.
One change is easy to miss. Sonnet 5 runs on an updated tokenizer, the same kind of change Anthropic made with Opus 4.7. The same input can now turn into roughly 1.0 to 1.35 times as many tokens, depending on what you send it. Anthropic says the introductory pricing is set to make that transition roughly cost-neutral, which is worth remembering once standard pricing starts on September 1.
Adaptive thinking is on by default, and the manual extended-thinking controls that existed on Sonnet 4.6 are gone; requests that try to set a thinking budget directly now return an error instead. If you are migrating an integration that depended on manual thinking controls, budget time to retest rather than assuming a drop-in swap.
Safety: What Changed
Anthropic's safety testing shows Sonnet 5 hallucinating and engaging in sycophantic behavior less often than Sonnet 4.6, and resisting prompt-injection hijack attempts more effectively. Cyber safeguards are on by default, the same detection system used on Opus 4.7 and 4.8, but Anthropic says it deliberately did not train Sonnet 5 on cybersecurity tasks. In one published test, built with Mozilla, that asked models to develop working exploits for already-patched Firefox vulnerabilities, Sonnet 5 never produced a complete exploit, though it showed a slightly higher partial-success rate than Sonnet 4.6. Both Sonnet models scored far behind Opus 4.8 and Mythos 5 on that test. Anthropic's own recommendation for security research that needs reduced guardrails is Opus 4.8, not Sonnet 5.
Benchmarks: How Close Is It to Opus 4.8?
Here is how Sonnet 5 compares on the evaluations Anthropic published at launch. SWE-bench Pro and Terminal-Bench 2.1 measure real coding and command-line agent work, Humanity's Last Exam tests graduate-level reasoning across disciplines, OSWorld-Verified scores how well a model operates a real computer interface, and GDPval-AA v2 grades performance on professional knowledge work.
| Benchmark | Sonnet 5 | Sonnet 4.6 | Opus 4.8 |
|---|---|---|---|
| Agentic coding (SWE-bench Pro) | 63.2% | 58.1% | 69.2% |
| Agentic coding (Terminal-Bench 2.1) | 80.4% | 67.0% | 82.7% |
| Reasoning, no tools (Humanity's Last Exam) | 43.2% | 34.6% | 49.8% |
| Reasoning, with tools (Humanity's Last Exam) | 57.4% | 46.8% | 57.9% |
| Computer use (OSWorld-Verified) | 81.2% | 78.5% | 83.4% |
| Knowledge work (GDPval-AA v2) | 1618 | 1395 | 1615 |
Figures are Anthropic's own published evaluation results as of June 30, 2026. On GDPval-AA v2, Sonnet 5 edges ahead of Opus 4.8, a narrow gap likely within methodology noise, but notable since it is the cheaper model.
On four of the five benchmarks, Sonnet 5 lands between Sonnet 4.6 and Opus 4.8, closer to Opus than to its own predecessor. On the fifth, it scores slightly higher than Opus 4.8: a genuine first for a mid-tier Sonnet model against its full-size sibling.
Pricing: Sonnet 5 vs the Rest of the Lineup
Price moves in the same direction as the benchmarks.
| Model | Input (per million tokens) | Output (per million tokens) |
|---|---|---|
| Sonnet 5 (intro, through Aug 31, 2026) | $2 | $10 |
| Sonnet 5 (standard, from Sep 1, 2026) | $3 | $15 |
| Sonnet 4.6 | $3 | $15 |
| Opus 4.8 | $5 | $25 |
| Fable 5 (currently suspended, for reference) | $10 | $50 |
Fable 5's pricing is listed for reference only. The model has been unavailable to the public since June 12, 2026, which is the subject of the next section.
The introductory rate holds through August 31, 2026. After that, Sonnet 5 settles at exactly what Sonnet 4.6 already costs, while undercutting Opus 4.8 by 40 percent on both input and output.
Why Sonnet 5 Is the Best Model Most People Will Actually Touch
Anthropic's own model lineup runs five tiers in ascending order of capability: Haiku, Sonnet, Opus, Fable, and Mythos. Sonnet 5 is the third tier from the top. Above it sits Opus 4.8, which remains fully available, and the Mythos-class models, Fable 5 and the unrestricted Mythos 5, which do not.
On June 12, 2026, the US Commerce Department sent Anthropic an export-control directive ordering it to suspend access to Fable 5 and Mythos 5 for any foreign national, anywhere, including Anthropic's own non-citizen employees. Anthropic could not technically separate foreign users from domestic ones fast enough to comply, so it disabled both models for every customer worldwide within hours. Every other Claude model, including Opus 4.8, kept running.
The two sides describe the trigger differently. The government's letter cited national security and a reported technique for bypassing Fable 5's safeguards, without giving Anthropic a detailed technical basis. Anthropic's public statement called the underlying jailbreak narrow and non-universal, said the same capability is available from other publicly deployed models, and said it believes the action is a misunderstanding. We are not adjudicating that dispute here; we track the live status, what changed, and what it costs you in our running Fable 5 timeline.
Two and a half weeks later, the suspension is still in force. The government cleared the more restricted Mythos 5 for a small set of vetted critical-infrastructure organizations on June 26, but that clearance did not extend to Fable 5, the version anyone with a subscription or API key was actually using. As of today, Anthropic says discussions with the administration that could restore access include the Sonnet 5 release itself, though no date is confirmed and earlier reports of an imminent return have already been retracted once. The timing carries its own stakes: Anthropic has confidentially filed for a US public listing, and a recent funding round valued the company near $965 billion, so how quickly this gets resolved is being watched well beyond Anthropic's own customers.
None of this changes what Sonnet 5 is. It changes what Sonnet 5 means. Opus 4.8 was already the highest-capability model available to the public before today. Sonnet 5 narrowing the gap to Opus, at a much lower price, is the more consequential development for most readers than anything happening above it, because the tier above it is not reachable right now regardless of price.
Should You Switch From Sonnet 4.6?
If you are on Sonnet 4.6 today, moving to Sonnet 5 is close to a free upgrade through August 31: better benchmarks across the board, at introductory pricing that lands at or below what you are already paying. After September 1, pricing returns to exactly Sonnet 4.6's old rate, so there is no ongoing cost to switch.
The one case for waiting is a workload tuned tightly around the old tokenizer's token counts, where the 1.0 to 1.35 times input change could matter for budget or context-window planning. Test a representative batch of your real requests before migrating anything running at scale.
If your actual target is Fable 5 or Mythos 5 rather than Opus-class performance, switching models will not get you there. That capability is gated by the export-control order, not by which Anthropic tier you choose.
If You'd Rather Not Depend on Any Cloud Model at All
Every model discussed so far runs on Anthropic's servers, under terms Anthropic sets and that a government agency can override in an afternoon, as Fable 5's customers learned on June 12. That is not a criticism specific to Anthropic. It is true of any closed, cloud-hosted model from any provider.
The alternative is hardware you own. Local models will not match Sonnet 5 or Opus 4.8 on the hardest agentic and reasoning work today, and pretending otherwise would not be honest. But for drafting, summarizing, document chat, and routine coding, open-weight models running on a $300 to $800 mini PC have closed most of the practical gap, and the terms of that setup cannot change underneath you by government order or by a pricing email. Our mini PC guide for local AI covers which hardware tier fits which workload, and our breakdown of what local models can actually do compares them directly against Claude's own benchmarks.
Frequently Asked Questions
Is Claude Sonnet 5 available now, and who can use it?
Yes, as of June 30, 2026. It is the default model for Free and Pro plans and is also available to Max, Team, and Enterprise users, in Claude Code, and through the Claude API as claude-sonnet-5.
How much does Claude Sonnet 5 cost?
$2 per million input tokens and $10 per million output tokens through August 31, 2026. Standard pricing after that is $3 per million input tokens and $15 per million output tokens, the same rate Sonnet 4.6 already charges.
Is Claude Sonnet 5 better than Claude Opus 4.8?
Not quite, on Anthropic's own published benchmarks, though it gets close on most of them and edges ahead on one knowledge-work test. Opus 4.8 remains the higher-accuracy option, at roughly 2.5 times the price.
Why isn't Claude Mythos 5 or Fable 5 available right now?
A US Commerce Department export-control order suspended both models worldwide on June 12, 2026, after a disputed report about a safeguard bypass. Anthropic disagrees with the order's basis and says it is working to restore access; no return date has been confirmed.
What is Claude Sonnet 5's context window?
1 million tokens, available at standard pricing with no beta header required, plus a 128,000 token output limit.
Should I switch from Sonnet 4.6, or wait for Fable 5 to come back?
Switch to Sonnet 5 now; it is a strict upgrade over 4.6 at the same eventual price. Waiting for Fable 5 solves a different problem. Even once it returns, it is a separate, more expensive, more restricted tier, not a replacement for your Sonnet workload.
