Claude Sonnet 5 Has Launched: What Actually Shipped

Back in February 2026, a version identifier — claude-sonnet-5@20260203 — leaked in Google Vertex AI logs, and speculation about “Claude 5” took off under the internal codename Fennec. That speculation is now settled: Claude Sonnet 5 officially launched on June 30, 2026.

This guide replaces the earlier leak-and-rumor coverage with what Anthropic actually confirmed at launch: pricing, benchmarks, features, and what it means if you’re building on Claude.

What We Know Now

Confirmed: The Model Is Live

Claude Sonnet 5 (model string claude-sonnet-5) launched on June 30, 2026, and replaced Sonnet 4.6 as the default model across Anthropic’s consumer and developer products the same day.

Confirmed: The Current Claude Lineup

Sonnet 5 sits between Haiku and Opus, as before — but it now performs close to where Opus 4.8 sits, at roughly 40–60% of the cost.

Note also that Anthropic’s frontier tier is now called Mythos, sitting above Opus. Fable 5 and Mythos 5 (released June 9, 2026) share the same underlying model, with Fable 5 carrying additional safety measures for biology, cybersecurity, and LLM R&D — which is why it’s the version available to the public while Mythos 5 isn’t.

Bonus: Universal Commerce Protocol (UCP): What Google’s AI Shopping Standard Means for Your Business

Confirmed: Pricing

Sonnet 5 launched with introductory pricing of $2 per million input tokens / $10 per million output tokens, in effect through August 31, 2026. After that, standard pricing kicks in at $3 per million input tokens / $15 per million output tokens — the same rate Sonnet 4.6 charged.

One catch worth knowing: Sonnet 5 uses a new tokenizer that encodes the same text into roughly 1.0–1.35x more tokens than Sonnet 4.6 did. Anthropic set the introductory pricing to make the transition roughly cost-neutral for now, but once standard pricing kicks in on September 1, effective costs on unchanged workloads could land 20–35% above where they started — even though the per-token rate card looks unchanged. If you’re budgeting for the switch, it’s worth re-testing your actual workloads rather than assuming a flat migration.

What Claude Sonnet 5 Actually Delivers

Agentic Performance

The clearest theme across Anthropic’s announcement and independent coverage: Sonnet 5 is built to be the most agentic Sonnet yet. It plans, uses tools like browsers and terminals, and runs autonomously at a level that previously required larger, more expensive models. Early testers reported it finishing multi-step jobs end-to-end (e.g., updating account records and sending follow-up communications) where earlier models would stall partway through.

On benchmarks reported by Anthropic:

• SWE-bench Verified: 72.7% (up from Sonnet 4.6’s 62.3%; Opus 4.8 sits at 79.4%)
• Terminal-bench: 76.1% (up from Sonnet 4.6’s 55.4% — the single largest jump in the release)

So the earlier prediction that Sonnet 5 would approach Opus-level performance at Sonnet pricing held up reasonably well — it’s not a full match, but it closes a meaningful part of the gap, and on some knowledge-work benchmarks it reportedly edges out Opus 4.8 entirely.

Context Window

The 1-million-token context window is now standard on Sonnet 5, not a beta feature — this is both the default and the maximum, with no smaller-context option. Keep the tokenizer change in mind here too: because tokens now cover slightly less text on average, the same million-token window holds somewhat less actual text than it did on Sonnet 4.6.

Thinking Behavior

Rather than a separate “zero-latency thinking” mode, what shipped is adaptive thinking, on by default. Manual extended thinking (the old approach of setting a fixed thinking budget) has been removed entirely on Sonnet 5, the same as on Opus 4.7 and 4.8 — you now control reasoning depth through an effort parameter instead.

“Dev Team Mode” — Partially Right

The most speculative leaked feature was a “Dev Team mode” where Claude would spawn specialized sub-agents to divide up complex work. That specific branded feature doesn’t appear to exist under that name. What did ship is real native multi-agent coordination and a model that’s noticeably better at sustained, multi-step autonomous work — so the underlying idea was directionally correct, even if the packaging wasn’t.

Safety

Anthropic reports Sonnet 5 shows an overall lower rate of undesirable behaviors than Sonnet 4.6, including improvements in rejecting malicious requests, resisting prompt-injection hijack attempts, and reducing hallucination and sycophancy. It’s also the first Sonnet-tier model with real-time cybersecurity safeguards built in — requests touching prohibited or high-risk cybersecurity topics can be refused outright.

Bonus: What is Claude Code And How To Use Claude Code (also Strenghts, Weakness and Pricing)

What This Means for Business Users

Content creation: Faster, cheaper access to near-Opus quality means more affordable iteration and higher-volume content production than was previously practical at this price point.

Customer support & automation: The real multi-step follow-through Sonnet 5 demonstrates (not a branded “Dev Team” feature, but genuine sustained agentic execution) is what actually matters here — it’s now more practical to hand off multi-part workflows without them stalling halfway.

Marketing & code workflows: Anthropic and early partners specifically called out gains in sustained coding, tool use, and debugging across messy technical contexts, plus meaningful upgrades in agentic search and computer-use tasks.

How Claude Sonnet 5 Compares to Competitors

The competitive picture has moved too. OpenAI’s GPT-5.6 Sol launched in preview around the same time with its own subagent-splitting approach. Google’s Gemini 3.5 Flash made a similar pitch, shifting from chatbot to autonomous planning tool. Sonnet 5 confirms agentic capability is now the baseline expectation for every lab’s mid-tier model.

On pricing, Sonnet 5 undercuts both Opus 4.8 and comparable higher-tier models from OpenAI and Google. It still costs more than Gemini’s cheapest Flash tier.

Bonus: ChatGPT vs Claude vs AI Marketing Platforms: Which Do You Need?

Claude Fable 5: The Frontier Tier Above Opus

Sonnet 5 wasn’t the only new model this cycle. Above Opus, Anthropic introduced a new tier called Mythos, and the first Mythos-class models — Claude Fable 5 and Claude Mythos 5 — launched on June 9, 2026, several weeks before Sonnet 5.

Fable 5 and Mythos 5 share the same underlying model. The difference lies in safety scope: Fable 5 carries additional safeguards around biology, cybersecurity, and LLM R&D, which is why Anthropic made it publicly available. Mythos 5 lacks that restriction layer, so Anthropic limits it to a small number of trusted organizations under Project Glasswing — the general public and standard API access can’t reach it. (Anthropic has more detail on Project Glasswing at anthropic.com/glasswing for anyone curious about that program.)

Pricing: Fable 5 is priced at $10 per million input tokens / $50 per million output tokens — the most expensive tier in the current lineup, well above Opus 4.8’s $5/$25.

A bumpy rollout: Fable 5 and Mythos 5’s launch wasn’t entirely smooth. On June 12, 2026 — just three days after release — Anthropic suspended access to both models to comply with U.S. Department of Commerce export controls. Those controls were lifted on June 30, and Anthropic restored access the next day, July 1, 2026. So for about two and a half weeks in mid-June, Fable 5 was effectively unavailable despite having just launched. Anthropic’s official statement on the suspension and restoration is available at anthropic.com/news/fable-mythos-access.

Fable 5 fits where maximum capability matters most, regardless of price, unlike Sonnet 5’s price-performance balance. The lineup now runs Haiku 4.5 to Sonnet 5 to Opus 4.8 to Fable 5. Mythos 5 sits alongside Fable 5, restricted to Project Glasswing partners only.

Should You Use Claude Sonnet 5 Now?

Claude Sonnet 5 has already shipped as the default model for Free and Pro plans. It’s also available on Max, Team, Enterprise, Claude Code, and the Claude API. If you’re already building on Claude 4.6, this upgrade path is straightforward, not a rebuild. Still, re-test your token consumption given the new tokenizer before assuming costs stay flat.

Key Takeaways

• Claude Sonnet 5 launched June 30, 2026 — the February leak was accurate, if a couple months early on timing.
• It’s the default model for Free and Pro plans, available across Max, Team, Enterprise, Claude Code, and the API.
• Pricing starts at $2/$10 per million tokens through Aug 31, then $3/$15 — but a new tokenizer may raise real costs.
• Performance: Closes much of the gap with Opus 4.8 on agentic and coding benchmarks, reportedly beating it on knowledge work.
• 1M-token context window is now standard, not beta.
• “Dev Team Mode” as a named feature didn’t materialize — but genuine multi-agent coordination and sustained autonomous execution did.
• Anthropic’s new Mythos tier sits above Opus — Fable 5 is public, while Mythos 5 stays limited to select partners.
• Anthropic launched Claude Fable 5 on June 9, 2026 at $10/$50 per million tokens, then briefly suspended it over export-control compliance.

This article reflects information available as of July 2026.

Sources:

• Introducing Claude Sonnet 5 — Anthropic
• Anthropic launches Claude Sonnet 5 as a cheaper way to run agents — TechCrunch
• What’s new in Claude Sonnet 5 — Claude Platform Docs

Building AI workflows for your business? Simplified AI Workflows let you create multi-agent automations today, with the flexibility to upgrade models as Claude 5 and future generations launch.