1. Advanced Coding Performance

• Achieves 74.5% on SWE-bench Verified, improving the Claude family's state-of-the-art coding abilities.

• Stronger at:

  • Multi-file code refactoring
  • Large codebase debugging
  • Pinpointing exact corrections without unnecessary edits

• Outperforms Opus 4 and shows gains comparable to jumps seen in past major releases.

2. Improved Agentic & Research Capabilities

• Better at maintaining detail accuracy in long research tasks.
• Enhanced agentic search and step-by-step problem solving.
• Performs reliably across complex multi-turn reasoning tasks.

3. Validated by Real-World Users

• GitHub: Better multi-file refactoring and code adjustments.
• Rakuten Group: High precision debugging with minimal collateral changes.
• Windsurf: One standard deviation improvement on their junior dev benchmark - similar magnitude to Sonnet 3.7 → Sonnet 4.

4. Hybrid-Reasoning Benchmark Improvements

• Improvements across TAU-bench, GPQA Diamond, MMMLU, MMMU, AIME (with extended thinking).
• Stronger robustness in long-context reasoning tasks.

1. Intelligence & Reasoning

• Outperforms previous Claude models and competitor LLMs across major benchmarks.
• Excels in graduate-level reasoning (GPQA), knowledge tasks (MMLU), and coding (HumanEval).
• Handles nuance, humor, and complex instructions with human-like clarity.

2. Speed & Efficiency

• Runs 2x faster than Claude 3 Opus, making it ideal for real-time and high-volume workflows.
• Cost-effective pricing: $3/M input tokens and $15/M output tokens.
• Supports a 200K token context window, enabling rich, long-form reasoning.

3. Coding Capabilities

• Solves significantly more coding and bug-fix tasks (64% vs Opus's 38% in internal evaluations).
• Can autonomously write, edit, and execute code when tool use is enabled.
• Strong at translating and modernizing legacy codebases.

4. Vision Strength

• Best vision model in the Claude family, surpassing Opus on vision benchmarks.
• Excellent at interpreting charts, graphs, and imperfect images.
• Reliable text extraction from low-quality visuals for retail, logistics, finance, etc.

5. Agentic Workflows

• Highly capable for multi-step task orchestration.
• Performs well as the engine for agents requiring reasoning, planning, and tool-calling abilities.

6. Content Quality

• Produces natural, relatable writing with improved tone, style, and context awareness.
• Strong at long-form content creation and editing.

7. Safety & Reliability

• Rated ASL-2, meeting Anthropic's safety standards.
• Undergoes extensive red-teaming and external evaluation (UK AISI & US AISI).
• Not trained on user data without explicit permission.