Anthropic’s Claude Sonnet 5, launched in late June, delivers near-Opus-level performance on coding and tool-use benchmarks while undercutting the flagship model’s price by more than half on both input and output tokens, according to benchmark data published by Anthropic and analyzed by MarkTechPost.
The mid-tier model achieves 63.2 percent on SWE-bench Pro, up from Sonnet 4.6’s 58.1 percent and approaching Opus 4.8’s 69.2 percent. On Terminal-Bench 2.1, Sonnet 5 scores 80.4 percent compared to Sonnet 4.6’s 67 percent. On OSWorld-Verified, a computer-use benchmark, it reaches 81.2 percent against 78.5 percent. On Humanity’s Last Exam with tools, both Sonnet 5 (57.4 percent) and Opus 4.8 (57.9 percent) are essentially tied.
On the knowledge-work benchmark GDPval-AA v2, Sonnet 5 actually edges ahead of Opus 4.8, scoring 1,618 versus 1,615.
Pricing is the differentiator. Through August 31, 2026, Sonnet 5 costs US$2 per million input tokens and US$10 per million output tokens. Standard pricing will settle at US$3 and US$15 respectively. Opus 4.8, by comparison, charges US$5 and US$25, roughly 67 percent more on both ends. The model also supports Anthropic’s 1-million-token context window.
However, the value proposition depends on task difficulty. At low and medium effort levels, Sonnet 5 delivers quality that earlier Sonnet pricing could not match. At the highest effort setting, its cost can exceed Opus 4.8 for similar quality, making the flagship model the better choice for accuracy-critical work. The new model uses a different tokenizer than Sonnet 4.6, the same as Opus 4.7, which can increase token counts by up to 35 percent, partially offsetting the per-token savings.
Early access partners including Zapier, Pace, and ClickHouse reported that Sonnet 5 handles multi-step software engineering workflows, writing reproducing tests, implementing fixes, and confirming their correctness in a single pass, as well as business automation tasks like updating Salesforce account tiers and sending end-to-end enterprise emails.
The practical routing strategy for developers: Sonnet 5 for the majority of agentic work, Opus 4.8 for the hardest tasks, and Haiku 4.5 for high-volume, latency-sensitive calls.
Sources: Anthropic Claude Sonnet 5 vs Sonnet 4.6 vs Opus 4.8: Agentic Coding Benchmarks, API Pricing, and Cost-Performance Tradeoffs Compared (MarkTechPost, July 2026); Anthropic official blog