claude sonnet 5 is the new default — 72.7% swe-bench, intro pricing through august

Anthropic released Claude Sonnet 5 on June 30, 2026, making it the default model for Free and Pro plan users in Claude.ai and Claude Code. the benchmark gains over Sonnet 4.6 are meaningful rather than incremental, and the pricing structure has a time-bounded intro window that's worth planning around.

coding benchmarks

on SWE-bench Verified — the standard real-world software engineering benchmark — Sonnet 5 scores 72.7%, up from 62.3% on Sonnet 4.6. that's a 10-point jump, which is substantial for a mid-tier model. for context: Claude Opus 4.8 scores 79.4% on the same benchmark, so Sonnet 5 is within 7 points of the Opus tier at Sonnet pricing.

on Terminal-bench, which measures performance on terminal-native agentic coding tasks — the kind Claude Code runs — Sonnet 5 scores 76.1%, up from 55.4% on Sonnet 4.6. a 20-point improvement on the benchmark most directly relevant to agentic coding workflows is the headline number here. Opus 4.8 sits at 79.4% on Terminal-bench, meaning Sonnet 5 is essentially at Opus performance on terminal tasks at roughly one-fifth the cost.

on SWE-bench Pro, a stricter version of the benchmark that filters for harder real-world issues: 63.2%.

pricing

the intro pricing window: $2 per million input tokens, $10 per million output tokens, through August 31, 2026. after that date, standard pricing is $3 input / $15 output.

for comparison: Sonnet 4.6 priced at $3/$15 standard. so Sonnet 5's intro pricing is lower than Sonnet 4.6's standard pricing with significantly better coding performance. after August 31, pricing is flat to Sonnet 4.6 with better benchmarks. if you're building on the API, the window to lock in lower costs on long-running workloads is now.

availability

Sonnet 5 is available in Claude (web and mobile), Claude Code, and the Claude Platform API. it's the default for Free and Pro plans — Max, Team, and Enterprise can also access it alongside other models. the model ID in the API is claude-sonnet-5-20260630 (check the Anthropic docs for the current recommended alias, as Anthropic pins aliases to specific checkpoint versions).

honest take

Sonnet 5 closing the gap with Opus 4.8 on terminal and agentic tasks matters more than it sounds. most Claude Code workflows run on Sonnet to save cost — if those workflows now match Opus performance on the benchmarks that measure their actual task type, the case for running Opus in those workflows gets weaker. Anthropic is compressing the performance gap between tiers, which is good for users and a pricing pressure on the Opus tier.

the August 31 intro pricing cutover is real. for teams running high-volume API workloads, the difference between $2 and $3 per million input tokens compounds quickly over two months of usage.