Claude Sonnet 5 Is Here — Near-Opus Performance at Sonnet Prices

On June 30, 2026, Anthropic released Claude Sonnet 5, its most capable mid-tier model to date — and the gap between "Sonnet" and "Opus" just got a lot smaller.

The headline claim is performance near-parity with Opus 4.8 at a fraction of the cost. On the SWE-bench Pro agentic coding benchmark, Sonnet 5 scores 63.2%, compared to Sonnet 4.6's 58.1% and Opus 4.8's 69.2%. For computer-use tasks measured by OSWorld-Verified, it hits 81.2% against Sonnet 4.6's 78.5%. On Terminal-Bench 2.1 — which tests long-horizon autonomous terminal work — Sonnet 5 leaps from 67.0% to 80.4%. On the knowledge-work benchmark GDPval-AA v2, it actually beats Opus 4.8: 1,618 vs 1,615.

The Agentic Difference

What makes Sonnet 5 stand out isn't a single benchmark score — it's how it handles chains of work. Anthropic designed it to plan sequences of tool calls, verify its own output without a human nudge, browse the web, execute terminal commands, and resume multi-step tasks without breaking down. In practice, you can hand it a codebase and an underspecified task and it will figure out the missing steps rather than stopping to ask.

Sonnet 4.6 required more scaffolding, more explicit instructions, and more fallback handling for agentic pipelines. Sonnet 5's self-checking behavior shifts some of that burden from the prompt to the model itself — it reviews its own outputs unprompted, catches common failure modes, and retries steps where it spots an inconsistency.

Pricing and Access

Introductory pricing runs through August 31, 2026: $2 per million input tokens and $10 per million output tokens. After that, it rises to $3/$15. Opus 4.8 costs considerably more, making Sonnet 5 the obvious default for most production agent workloads that don't require absolute peak capability.

The model is already the default for Free and Pro Claude.ai plans. Developers can reach it via the Claude API using the model ID claude-sonnet-5.

Why This Changes the Calculus

The trend in AI deployment is toward agentic use cases — models that don't just respond to a single prompt but manage long tasks autonomously. The bottleneck has been that the models capable of sustaining multi-step reasoning were expensive enough to make production deployments prohibitive at scale.

Sonnet 5 shifts that equation. A model that matches Opus on knowledge tasks, comes close on coding, and costs two-thirds less per token changes what's practical to deploy. For developers building on top of Claude, the upgrade path from 4.6 is essentially free in terms of effort — same API, better output, less error-correction scaffolding to build around the model.

As TechCrunch noted at launch, the model represents Anthropic's clearest statement yet that agentic AI should be affordable, not a premium tier reserved for the most well-funded deployments.