Anthropic Launches Claude Sonnet 4.5 — Best Coding Model, 30-Hour Sustained Focus
The 30-hour sustained focus capability moves AI agents from controlled demos into genuine operational infrastructure — making it worth revisiting any AI initiative your team shelved because agents weren't reliable enough for long-horizon tasks.
Anthropic Launches Claude Sonnet 4.5 — Best Coding Model, 30-Hour Sustained Focus
By Anthropic
What's New
Anthropic has released Claude Sonnet 4.5, its most capable Sonnet model to date. The release marks a meaningful step forward in autonomous AI performance — particularly for organizations already running or planning to run agentic workflows at scale.
Benchmark Performance
Sonnet 4.5 sets a new bar across several key dimensions:
- Coding: 77.2% on SWE-bench Verified, making it Anthropic's strongest coding model and among the best available
- Computer use: 61.4% on OSWorld — up sharply from 42.2% — indicating significantly improved ability to operate software interfaces autonomously
- Sustained focus: Maintains coherent task execution across multi-step workflows for 30+ hours without degradation
- Domain expertise: Dramatically improved performance in finance, law, medicine, and STEM
Why the 30-Hour Focus Window Matters
The headline capability for enterprise leaders is not the benchmark scores — it's the sustained attention span. Previous models could handle short bursts of autonomous work, but dropped context or made compounding errors over longer horizons. A 30-hour sustained focus window means real autonomous agent workloads are now viable: overnight research tasks, multi-day due diligence pipelines, continuous code review loops, and complex multi-tool orchestration that couldn't be reliably delegated before.
This shifts AI agents from impressive demos into operational infrastructure.
Pricing
Pricing remains unchanged from Sonnet 4:
- Input: $3 per million tokens
- Output: $15 per million tokens
More capability at the same price point improves the unit economics of any AI workflow built on this model.
Availability
Claude Sonnet 4.5 is available now via the Anthropic API and on Claude.ai.
The Bottom Line for Leaders
Sonnet 4.5 is not an incremental polish — it's the model that makes long-horizon autonomous agents a credible operational bet. If your organization has been waiting for AI agents that can reliably complete complex, multi-step tasks without human intervention at each checkpoint, this is the capability threshold worth re-evaluating against your current AI roadmap.