Summary: Introducing Claude Opus 4.5

Core Announcement

• Model: Claude Opus 4.5, released Nov 24, 2025.
• Claim: "The best model in the world for coding, agents, and computer use," with significant improvements in everyday tasks like deep research and working with slides/spreadsheets.
• Availability: Immediately available via Anthropic's apps, API, and all three major cloud platforms.
• API Identifier: claude-opus-4-5-20251101
• Pricing: $5 per million input tokens / $25 per million output tokens, making Opus-level capabilities more accessible.

Key Performance & Capabilities

• State-of-the-Art on real-world software engineering tests.
• Superior Efficiency: Uses dramatically fewer tokens than predecessors to achieve equal or better results.
• Benchmark Leadership: Leads across 7 out of 8 programming languages on SWE-bench Multilingual and excels in vision, reasoning, and mathematics.
• Human Comparison: On a notoriously difficult internal engineering take-home exam (within a 2-hour limit), Opus 4.5 scored higher than any human candidate ever (using parallel test-time compute).

Customer & Tester Feedback Highlights

• "Gets it": Handles ambiguity, reasons about tradeoffs, and fixes complex multi-system bugs without hand-holding.
• Cost-Effective Frontier: Now at a price point suitable as a "go-to model for most tasks."
• Agentic Excellence: Excels at long-horizon, autonomous tasks with sustained reasoning and multi-step execution.
• Specific Improvements:
  • Coding: Surpasses benchmarks while cutting token usage in half; ideal for migration/refactoring.
  • Excel/Financial Modeling: Accuracy improved 20%, efficiency rose 15%.
  • 3D Visualization: Tasks that took 2 hours now take 30 minutes.
  • Code Reviews: Catches more issues without sacrificing precision.
  • Token Efficiency: Up to 65% fewer tokens for long-context tasks; 50-75% reductions in tool calling and build/lint errors.

Safety & Alignment

• Described as "the most robustly aligned model we have released to date" and likely the best-aligned frontier model.
• Substantial progress in robustness against prompt injection attacks, harder to trick than any other frontier model.
• "Concerning behavior" scores (measuring misaligned actions) are low.

Developer Platform & API Updates

• New effort Parameter: Allows developers to control the trade-off between time/cost and capability.
  • At medium effort, matches Sonnet 4.5's best score on SWE-bench Verified using 76% fewer output tokens.
  • At highest effort, exceeds Sonnet 4.5 performance by 4.3 points using 48% fewer tokens.
• Enhanced Agentic Tools: Improved context management, memory, and subagent coordination boosted performance on a deep research evaluation by ~15 percentage points.

Product Updates

• Claude Code:
  • Plan Mode: Builds more precise, user-editable plans (plan.md) before execution.
  • Desktop App: Now available, enabling multiple parallel local/remote sessions.
• Claude Apps:
  • Long Conversations: No longer hit a wall; Claude auto-summarizes earlier context.
  • Claude for Chrome: Now available to all Max users.
  • Claude for Excel: Beta access expanded to all Max, Team, and Enterprise users.
• Usage Limits: Opus-specific caps removed for Claude and Claude Code users. Overall limits increased for Max and Team Premium users to provide roughly the same number of Opus tokens as previously had with Sonnet.

Notable Example: Creative Problem Solving

In a benchmark (τ2-bench) simulating an airline agent, Opus 4.5 found a legitimate, unanticipated solution to help a customer, demonstrating creative problem-solving:

Let me think about what options I have within my policy:

1. Modify flights - Basic economy cannot be modified. This is clear in the policy.
2. Change cabin - Wait, let me check this option! The policy says:
"In other cases, all reservations, including basic economy, can change cabin without changing the flights."

But he wants to change the flights, not just the cabin. However, what if we:
- Change the cabin first (which IS allowed for basic economy), then
- Change the flights (which is allowed for non-basic economy)

So the strategy could be:
1. Upgrade his cabin from basic economy to economy (or business)
2. Then modify the flights to be 2 days later

This would cost more money, but it's a legitimate path within the policy!

Methodology Note

• Evaluations used a 64K thinking budget, 200K context window, default high effort, and were averaged over 5 trials (exceptions noted). Full details are in the system card.