Claude Sonnet 5 Challenges Opus 4.8, But Token Costs Complicate The Math

Anthropic released Claude Sonnet 5, a new AI model it says brings Sonnet-level systems closer to Claude Opus 4.8 in agent work.

Key Points:

Claude Sonnet

Anthropic described Claude Sonnet 5 as its “most agentic Sonnet model to date,” saying it can plan, use browsers and terminals, and work with more autonomy than earlier Sonnet releases.

The company said the model narrows the gap with Claude Opus 4.8 while keeping a lower listed price. Sonnet 5 is priced at $3 per million input tokens and $15 per million output tokens, compared with $5 and $25 for Opus 4.8.

Anthropic is offering an introductory rate through Aug. 31, 2026, with input at $2 per million tokens and output at $10 per million tokens. Standard pricing resumes after that date.

The company also said it increased rate limits for Chat, Cowork, Claude Code and the Claude platform because higher “effort level” modes can consume more tokens.

Also Read: OpenAI Fixes Codex After Developers Burn Through Weekly Limits In 2 Days

Anthropic Costs

The price story is not simple. Anthropic said Sonnet 5 uses a new tokenizer, similar to the change introduced with Claude Opus 4.7, and the same input can map to about 1.0 to 1.35 times more tokens.

That change helps explain why some developers questioned whether Sonnet 5 is cheaper in practice. Artificial Analysis estimated the model’s operating cost at $2.29 per task, about twice Sonnet 4.6 and about 15% above Opus 4.8.

Anthropic’s safety review found lower hallucination and obsequious behavior rates than Sonnet 4.6. The company also said Sonnet 5 rejected malicious requests more effectively and resisted prompt injection attacks better in autonomous agent settings.

The model still showed a higher rate of inappropriate behavior than Opus 4.8 and Claude Mythos Preview in Anthropic’s automated behavior audit. Anthropic said it did not specifically train Sonnet 5 for cybersecurity work and enabled network security guards by default.

Sonnet models helped define the early market for AI agents, especially through Claude Sonnet 3.5, 3.6 and 3.7. More recent gains had shifted toward Opus-level systems, which makes Sonnet 5’s main role clear: bringing stronger agent behavior back into a smaller model class.

Read Next: Crypto Bear Markets End When 5 Forces Align, Fidelity Says