Anthropic made Claude Sonnet 5 the default model for every user on June 30 at $2/$10 per million tokens. The mid-tier model performs at Opus 4.8 levels on roughly 80% of agentic coding tasks at half the cost. A new self-hosted code gateway opens regulated industries. The strategy signals Anthropic is betting volume replaces margin faster than expected.
Anthropic made Claude Sonnet 5 the default model for every Free and Pro user on June 30, and the move tells you more about the state of the AI industry than any benchmark table. The top-tier Opus model is no longer the default. The mid-tier Sonnet is. And the pricing suggests Anthropic is betting that volume replaces margin faster than anyone expected.
Claude Sonnet 5 shipped with an introductory price of $2 per million input tokens and $10 per million output — half of Opus 4.8's pricing and comparable to GPT-5.6 Terra. After August 31, the price steps to $3/$15, still below the Opus tier. The model runs on the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Azure. It's the default across all Anthropic subscription tiers: Free, Pro, Max, Team, and Enterprise.
The positioning is explicit. Sonnet 5 isn't a downgrade from Opus. It's a redefinition of what "good enough" means for production workloads. Anthropic is telling developers: you don't need Opus for most tasks. Sonnet 5 is strong enough for agentic coding, multi-step reasoning, and production automation — and it costs half as much.
The coding benchmarks back this up. Sonnet 5 scores materially higher than Sonnet 4.6 on SWE-Bench, HumanEval, and MBPP. On agentic coding tasks — multi-file refactoring, debugging complex systems, writing tests from specifications — it reportedly performs at Opus 4.8 levels on roughly 80% of tasks. The 20% gap is real, but for most production pipelines, an 80% solution at 50% cost is the better business decision.
Anthropic also launched a self-hosted code gateway alongside Sonnet 5 — a deployment option that lets enterprises run the model on their own infrastructure rather than routing through Anthropic's cloud. This matters for regulated industries. Healthcare, finance, defense, and government customers with data residency requirements can now deploy Sonnet 5 without sending code or data to a third-party API endpoint. It's a direct shot at OpenAI's cloud-only deployment model.
The enterprise impact is already visible. Anthropic reportedly closed several large contracts in the two weeks following the launch, including deals in financial services and legal tech where the self-hosted gateway was the deciding factor. Companies that couldn't use frontier AI because of compliance requirements now have a path — and Anthropic is the only major lab offering that path at this performance level.
The broader strategic signal is even more important. Anthropic is compressing its product hierarchy. Instead of "Opus for hard problems, Sonnet for medium problems, Haiku for easy problems," the company is moving toward "Sonnet for everything, Opus for the hardest 20%, Haiku for latency-sensitive tasks." That's a volume play. The margin on Opus is higher, but the addressable market for Sonnet is an order of magnitude larger.
The introductory pricing through August 31 creates urgency. Developers who build on Sonnet 5 during the discount window face a price increase if they stay, but switching costs after integration reduce the likelihood they'll leave. It's a customer acquisition strategy disguised as a promotion — and given Anthropic's revenue lead over OpenAI ($47B annualized vs. $25-33B), it appears to be working. One data point deserves more attention than it's getting: Anthropic made Sonnet 5 the default for Free users. Not a preview. Not a limited trial. The default. That means anyone with a web browser can access a model that competes with last year's frontier systems — for free. The gap between what the richest companies can afford and what anyone can access has narrowed dramatically in six months. Sonnet 5 is both cause and evidence of that shift.
Get AI news in your inbox
Daily digest of what matters in AI.
Key Terms Explained #
Anthropic An AI safety company founded in 2021 by former OpenAI researchers, including Dario and Daniela Amodei.
Attention A mechanism that lets neural networks focus on the most relevant parts of their input when producing output.
Benchmark A standardized test used to measure and compare AI model performance.
Claude Anthropic's family of AI assistants, including Claude Haiku, Sonnet, and Opus.