Claude Sonnet 5 Closes the Gap With Opus at a Fraction of the Cost

Anthropic released Claude Sonnet 5 on June 30, positioning it as a cheaper way to run agentic workloads that previously needed the larger Opus 4.8 model. It became the default model for Free and Pro plans at launch and is also live on Max, Team, and Enterprise plans, the API, AWS Bedrock, and Google Cloud Vertex AI.

The pitch is near-Opus performance on planning, tool use, and autonomous multi-step work at a lower price. Sonnet 5 actually edges past Opus 4.8 on two of six benchmarks Anthropic published — Terminal-Bench 2.1, where it jumped over 13 points generationally, and a knowledge-work benchmark called GDPval-AA v2. On the other four, including SWE-bench Pro and Humanity's Last Exam, Opus 4.8 stays ahead, though the gap has narrowed substantially versus Sonnet 4.6.

Pricing launched at $2 per million input tokens and $10 per million output tokens through August 31, rising to $3/$15 after that — still undercutting Opus 4.8, GPT-5.5, and Gemini 3.1 Pro. Anthropic also reported lower rates of undesirable behavior (deception, sycophancy, misuse cooperation) than Sonnet 4.6, alongside better prompt-injection resistance, while noting Sonnet 5 sits below Opus-tier models on misalignment-related safety evaluations.

The release fits a broader industry pattern this summer: agentic capability that once required flagship-tier models is becoming standard at the mid-tier, with OpenAI's [GPT-5.6 Sol](/models/gpt-5-6-sol) and Google's Gemini 3.5 Flash making similar moves in recent months. Note: the benchmark figures above come from Anthropic's own launch materials and haven't yet been independently reproduced.