Changelog

Opus 4.5 is now available in Augment, and we're giving it to you at Sonnet prices for the next two weeks.

Why it matters

In our internal evaluations, we observed that developers increasingly rely on reasoning-first models for tasks involving multi-file coordination, implementation planning, and structured decision-making. Opus 4.5 is designed to meet that demand.

What improved

• Stronger reasoning capabilities Better handling of multi-step implementation details and edge case considerations.
• Improved tool call efficiency Reaches the correct action more consistently, often with fewer back-and-forth steps.
• Fewer thinking tokens than other reasoning models One downside of reasoning models is that their thinking blocks can be verbose, which may lead to higher perceived latency. From our experience with Opus 4.5, we’ve observed a significant improvement in latency by using thinking tokens more efficiently per turn.
• More focused and direct responses Avoids unnecessary narration on obvious edits; explanations tend to be clearer and more concise.

What to expect

• Occasional delayed time-to-first-token when processing extensive code context.
• Some asynchronous workloads showed intermittent slowdowns, currently under evaluation.

When to use it

Best suited for:

• Multi-step refactors
• Feature planning / reasoning
• Tasks requiring structured breakdown
• Agent-based workflows

Less ideal for:

• Quick or trivial edits
• Speed-critical workflows
• Very high concurrency

We optimized Opus 4.5 for depth and clarity over raw speed on trivial edits. We think it could become your daily driver. Try it today and let us know what you think!