Claude Sonnet 5 is Anthropic's newest model, and it's built specifically to tame the agentic AI workloads that have been driving up bills for enterprise customers and heavy users of AI assistants. The company says Sonnet 5 delivers performance close to its flagship Opus models while costing significantly less to run, largely thanks to a redesigned tokenizer that makes every request more efficient.
What's new in Claude Sonnet 5
Unlike earlier Claude releases aimed mainly at answering direct human questions, Sonnet 5 was tuned around the reality that agentic tools now generate far more queries per session than any person could type manually. An AI agent might call a model dozens or hundreds of times to complete a single task, searching, reasoning, calling other tools, and checking its own work, and each of those calls adds up. Anthropic says Sonnet 5 matches the performance of its recent Opus models on these agentic benchmarks, but does so more efficiently, which is the difference that shows up on a company's monthly invoice.
Claude Sonnet 5 pricing and availability
Starting September 1, Sonnet 5 will cost $3 per million input tokens and $15 per million output tokens as a base rate. Anthropic says prices will be even lower than that before the September date arrives. For comparison, Opus 4.8 currently costs $4 per million input tokens and $25 per million output tokens. Sonnet 5 is available now in Claude Code and on the Claude Platform at these rates, and it's also rolling out as the default model across Anthropic's consumer plans, including the free and Pro tiers.
Why agentic AI is running up enterprise bills
Agentic AI tools are designed to work with minimal supervision, chaining together searches, code execution, and file edits until a task is done. That autonomy is the whole selling point, but it also means token consumption scales in ways that are hard to predict. A single automated workflow can rack up the same volume of requests a human might generate in weeks, and businesses that lean on agents for coding, research, or customer support have found their bills climbing accordingly.
The problem is structural, not just a matter of heavy users being careless. A human typing a question into a chatbot sends one request and reads one answer. An agent working through a coding task, by contrast, might read a file, run a test, check the output, adjust its approach, and repeat that loop dozens of times before finishing, with every one of those steps consuming input and output tokens. Multiply that across a team running several agents at once, and the token bill can outpace what any per-seat software pricing model ever anticipated.
What Sonnet 5 means for businesses running AI agents
By pairing near-Opus performance with a cheaper, more efficient tokenizer, Anthropic is betting that Sonnet 5 becomes the default engine for agentic products rather than a downgrade users pick only to save money. That matters for any company building AI-powered products, since it lowers the cost floor for running always-on agents without necessarily sacrificing capability. It won't eliminate runaway spending entirely, a poorly designed agent can still burn through tokens quickly, but it does give teams a cheaper lever to pull before reaching for the more expensive Opus tier.
FAQ
How much does Claude Sonnet 5 cost?
From September 1, Sonnet 5 costs $3 per million input tokens and $15 per million output tokens, with lower promotional pricing available before that date.
Is Claude Sonnet 5 available for free users?
Yes. Sonnet 5 is rolling out as the default model across all Anthropic subscription tiers, including the free and Pro plans, not just paid enterprise accounts.
How is Sonnet 5 different from Opus 4.8?
Anthropic says Sonnet 5 offers similar performance on agentic tasks to Opus 4.8 but at a lower price, $3/$15 per million input/output tokens versus $4/$25 for Opus 4.8.
