Claude Opus 4.5 scores 80.9% on SWE-bench, cuts costs 66% with $5/$25 pricing after $350B Anthropic valuation

Anthropic released Claude Opus 4.5 on November 24-25, 2025, positioning it as the most capable model for agentic coding and computer use. The release follows rapid succession launches across the Claude 4.5 family, with Sonnet 4.5 arriving in late September and Haiku 4.5 in mid-October 2025.

The model achieves 80.9% accuracy on SWE-bench Verified, a benchmark measuring real-world software engineering tasks. This performance exceeds OpenAI’s GPT-5.1-Codex-Max at 77.9%, Anthropic’s own Sonnet 4.5 at 77.2%, and Google’s Gemini 3 Pro at 76.2%. Opus 4.5 became the first model to surpass the 80% threshold on this benchmark, establishing new performance standards for autonomous coding capabilities.

Pricing dropped significantly from the previous Opus generation. The new model costs $5 per million input tokens and $25 per million output tokens, down from $15 and $75 respectively. This 66% price reduction makes frontier-level capabilities accessible to broader developer audiences while maintaining competitive positioning against Google’s Gemini models and OpenAI’s GPT family.

The strategic partnerships announced November 18, 2025, reshaped Anthropic’s infrastructure capabilities. Microsoft committed to invest up to $5 billion while Nvidia pledged up to $10 billion, pushing the company’s valuation to approximately $350 billion from $183 billion in September. Anthropic committed to purchasing $30 billion of Azure compute capacity and up to one gigawatt of additional capacity for model training.

Anthropic tested the model on its internal performance engineering exam given to prospective hires. Using parallel test-time compute within the two-hour limit, Opus 4.5 scored higher than any human candidate who has taken the assessment. The company noted this result doesn’t measure collaboration, communication, or professional instincts that develop through experience, but it demonstrates technical problem-solving capabilities under time constraints.

The model introduces an effort parameter through the Claude API, allowing developers to balance computational work against latency and cost. At medium effort, Opus 4.5 matches Sonnet 4.5’s performance on SWE-bench Verified while using 76% fewer output tokens. At maximum effort, it exceeds Sonnet 4.5 by 4.3 percentage points while still consuming 48% fewer tokens, according to Anthropic’s technical documentation.