Anthropic’s Claude Sonnet 5 Closes Gap With Opus, Fable 5 Returns

Anthropic released Claude Sonnet 5 on June 30, 2026, with introductory API pricing that puts near-Opus performance inside a Sonnet budget. The model is the default for Free and Pro subscribers, available to Max, Team, and Enterprise users, and ships at $2 per million input tokens and $10 per million output tokens through August 31, 2026.

The same day, the US Commerce Department lifted export controls on Claude Fable 5 and Mythos 5, ending an 18-day standoff that had shut both models off to every customer worldwide. Fable 5 returns to Claude.ai, Claude Code, and the Claude Platform on July 1, with global access on AWS, Google Cloud, and Microsoft Foundry to follow, Anthropic said.

How Sonnet 5 Is Priced

Anthropic opened Sonnet 5 to every subscription tier on day one and set a temporary rate that holds through August 31, 2026. After that window closes, the model moves to standard pricing of $3 per million input tokens and $15 per million output tokens. The introductory discount matches the window Anthropic used on earlier Sonnet launches, and it lands in a market where agentic capability has been treated as a premium feature. Per Anthropic’s Sonnet 5 launch announcement, the company raised rate limits across Chat, Cowork, Claude Code, and the Claude Platform to absorb the higher token usage of bigger effort levels.

Sonnet 5 sits one tier below Opus 4.8 in Anthropic’s lineup and is being pitched as the agentic workhorse for most customer tasks. Anthropic’s launch post says the model’s performance is “close to that of Opus 4.8, but at lower prices.” Launch-day coverage reports Sonnet 5 is now cheaper than Opus 4.8, OpenAI’s GPT-5.5, and Google’s Gemini 3.1 Pro, while still costing more than Google’s Gemini 3.5 Flash.

Anthropic also raised rate limits across Chat, Cowork, Claude Code, and the Claude Platform to absorb higher token usage at bigger effort levels, and developers can pull Sonnet 5 through the Claude API under the model name claude-sonnet-5. The table below lays out the introductory and standard rates as Anthropic published them, side by side.

What the New Model Does Differently

Sonnet 5 is the most agentic Sonnet model Anthropic has shipped, a label the company applies in its own launch post. It can plan multi-step tasks, drive browsers and terminals as tools, and self-verify its own output without being prompted, the kinds of skills that previously sat in the Opus tier. The model runs on a million-token context window with up to 128,000 tokens of output per response, an expansion that matters for agents working across large codebases. Anthropic also launched Sonnet 5 with cyber safeguards enabled by default and a lower rate of cooperation with misuse than its predecessor. The behavior changes, including adaptive thinking on by default and the removal of manual extended thinking, are detailed in Anthropic’s Sonnet 5 developer notes.

Early-access partners said Sonnet 5 finishes what previous Sonnets left half-done. Zapier senior engineer Daniel Shepard said his team handed the model a two-part job that paired Salesforce updates with a launch announcement. “That used to stall halfway,” Shepard said. Lovable co-founder Fabian Hedin said in a statement that refusal behavior is what makes Sonnet 5 usable at scale.

At Lovable, we’re putting powerful tools in the hands of millions of builders. A model that knows when to say no is just as important as one that knows how to build.

Where Sonnet 5 Stands Against Opus and Rivals

Anthropic’s benchmark chart puts Sonnet 5 within striking distance of Opus 4.8 on agentic coding. On one evaluation, Sonnet 5 scored 63.2%, against Opus 4.8’s 69.2% and the previous Sonnet 4.6’s 58.1%. Anthropic’s launch post says higher-effort Sonnet 5 calls “can match Opus 4.8 on some tasks.” The full evaluation set is in Anthropic’s Sonnet 5 system card PDF.

The gap is small enough that Anthropic is now telling customers to pick a tier based on effort level, not raw capability.

Launch-day coverage also reports Sonnet 5 is now priced below OpenAI’s GPT-5.5 and Google’s Gemini 3.1 Pro, though it remains more expensive than Google’s Gemini 3.5 Flash, the budget-priced model Google pushed in May. Anthropic frames the comparison as cost-adjusted performance: between Sonnet 5 and Opus 4.8, “users can adjust the effort level to find the right balance of cost and performance.”

Independent testing surfaces a wrinkle in the price story. Sonnet 5 ships with an updated tokenizer that maps the same input text to roughly 1.0 to 1.35 times as many tokens as Sonnet 4.6, depending on content. Anthropic calls the change an intentional tradeoff for performance and says the introductory discount is set so the transition is roughly cost-neutral. A user moving a million-token English document from Sonnet 4.6 to Sonnet 5 is likely to be billed for around 1.4 times as many tokens, even before the standard rate kicks in on September 1.

How the 18-Day Export Standoff Ended

Anthropic launched Fable 5 and Mythos 5 on June 9, 2026, the first publicly available model in the Mythos tier. Three days later, on June 12, the Commerce Department sent Anthropic a directive ordering it to suspend access for any foreign national, including the company’s own non-citizen staff. Anthropic said it could not verify nationality in real time, so it disabled both models globally within hours. The earlier partial clearance for Mythos 5, handed down on June 26 to roughly 100 US companies and federal agencies, is detailed in how the partial Mythos clearance unfolded.

The freeze hit every customer at once. The directive forced Anthropic to pull the models from AWS Bedrock, Google Cloud, and Microsoft Foundry alongside its direct Claude APIs, with no exception for any enterprise client. The dispute centered on a threat-intelligence report from Amazon that described a technique for bypassing one of Fable 5’s cybersecurity safeguards, and further reporting found that equivalent and lesser models could produce the same output.

On June 30, 2026, Commerce Secretary Howard Lutnick posted on X that the controls were lifted, and Anthropic confirmed the move the same day. Forbes calls the freeze an 18-day standoff. Fable 5 returns to Claude.ai, Claude Code, Claude Cowork, and the Claude Platform on July 1, with access on AWS, Google Cloud, and Microsoft Foundry to follow “as quickly as possible.”

• June 9, 2026: Anthropic releases Fable 5 publicly and Mythos 5 to Project Glasswing partners.
• June 12, 2026: Commerce Department issues the export-control directive.
• June 13, 2026: Anthropic disables both models for every customer.
• June 26, 2026: Lutnick letter partially clears Mythos 5 for roughly 100 trusted US partners.
• June 30, 2026: Export controls fully lifted; Fable 5 restoration set for July 1.

What Lutnick’s Letter Actually Permits

The letter Lutnick sent, addressed to Anthropic co-founder Tom Brown rather than CEO Dario Amodei, applies to both frontier models. “A license is no longer required for the export, reexport, or in-country transfer, including deemed export or deemed reexport, of the Mythos or Fable models,” Lutnick wrote.

Anthropic has agreed to detect and address security risks in the models proactively, work with the US government on protocols for future releases, and notify the government if it sees malicious activity tied to either model. The Lutnick letter also reserves the right for the Trump administration to reimpose restrictions if circumstances change or if Anthropic falls short on those commitments. Mythos 5, the more capable of the two, had already been partially restored to about 100 US companies and federal agencies under the earlier June 26 letter.

The OpenAI Contrast on the Same Week

OpenAI launched GPT-5.6 in three variants, Sol, Terra, and Luna, on June 26, the same day Anthropic’s partial clearance landed. The release was previewed to the US government ahead of time, and at the government’s request OpenAI initially made the model available only to a small group of trusted partners whose participation had been shared with officials. OpenAI said it would keep coordinating with government partners before expanding access.

The company also pushed back on the precedent it was setting. “We do not believe this kind of government access process should become the long-term default,” OpenAI said, because it keeps advanced AI tools from “users, developers, enterprises, cyber defenders, and global partners who need them.”

The Office of the National Cyber Director and the Office of Science and Technology Policy were both involved in the request, and CEO Sam Altman told staff that access would be approved customer by customer. That hands federal officials a gatekeeping role over early deployments of the model and could slow enterprise adoption timelines.

Anthropic’s June 30 announcement lands in the same regulatory frame. Sonnet 5 sits well below the cyber-capability threshold that triggered the Fable 5 directive, and Anthropic shipped it with cyber safeguards enabled by default. Anthropic’s blog post argues that government involvement in AI releases “requires a durable, transparent process” codified in “strong regulation and applied equally across frontier model developers.”

What Enterprise Buyers Should Watch

For enterprise buyers, the deeper lesson is structural. “Frontier access has become conditional infrastructure,” said Sanchit Vir Gogia, chief analyst at Greyhound Research. “The models went dark globally because nationality could not be verified in real time, so a control aimed at foreign nationals became an outage for everyone.” Enterprises that mapped their model dependencies ahead of time could hot-swap to Claude Opus 4.8 within hours. Those that did not, Forbes reports, discovered that force majeure clauses were never written with a government-mandated AI cutoff in mind.

Buying across multiple cloud providers does not insulate a buyer from action against the underlying model provider, Gogia told CIO.com. Restoration timing, he says, is now a procurement criterion alongside traditional cost, security, and performance considerations, and “restored access is not restored certainty.”

Anthropic used the announcement to call for an industry-wide framework for evaluating AI jailbreaks, saying developers and governments lack a common standard for severity. The company says it is working with Amazon, Microsoft, Google, and other Project Glasswing partners on that framework, while expanding pre-release testing and information sharing with the US government.

Frequently Asked Questions

How much does Claude Sonnet 5 cost?

Anthropic is charging $2 per million input tokens and $10 per million output tokens through August 31, 2026, the introductory window for the launch. After that date, the standard rate of $3 per million input tokens and $15 per million output tokens takes effect, returning to the same per-token prices Sonnet 4.6 carried.

How does Sonnet 5 compare to Opus 4.8?

Sonnet 5 is positioned as the cheaper, more agentic option one tier below Opus 4.8. Anthropic’s launch post says the model’s performance is close to Opus 4.8 at lower prices, with Sonnet 5 scoring 63.2% on one agentic coding benchmark compared to Opus 4.8’s 69.2% and matching Opus on some tasks at higher effort levels.

When will Fable 5 access be restored?

Anthropic says Fable 5 returns to Claude.ai, Claude Code, Claude Cowork, and the Claude Platform on July 1, 2026, the day after the export controls were lifted. Access on AWS, Google Cloud, and Microsoft Foundry will follow on a rolling basis as the cloud providers re-provision the model.

Why were Fable 5 and Mythos 5 blocked?

The Commerce Department issued a June 12, 2026 directive after a threat-intelligence report from Amazon described a technique for bypassing one of Fable 5’s cybersecurity safeguards. Anthropic has since retrained its safety classifier and says the reported technique is now blocked in more than 99% of cases, while acknowledging the stronger safeguards will flag more legitimate coding requests.

How does this compare to OpenAI’s GPT-5.6 launch?

OpenAI launched GPT-5.6 Sol, Terra, and Luna on June 26, 2026, with a limited trusted-partner rollout at the US government’s request and customer-by-customer approval from the Office of the National Cyber Director. Anthropic’s Sonnet 5 ships more broadly, with availability to Free, Pro, Max, Team, and Enterprise users on day one, well below the cyber-capability threshold that triggered the Fable 5 directive.