**Anthropic has released Claude Sonnet 5, the most capable mid-tier model the company has ever built, and made it the default for every Free and Pro user starting today, July 1, 2026.**
The launch marks a strategic pivot: rather than competing purely on raw capability, Anthropic is betting that the market's next frontier is agentic performance at accessible price points. Sonnet 5 delivers capabilities that, just months ago, required the much larger and more expensive Opus 4.8.
"It can make plans, use tools like browsers and terminals, and run autonomously at a level that, just a few months ago, required larger and more expensive models," Anthropic said.
Key benchmarks reveal how close Sonnet 5 is to flagship performance:
- Agentic Coding: 63.2% (vs Opus 4.8 at 69.2%, Sonnet 4.6 at 58.1%) - OSWorld-Verified (desktop automation): 81.2% (vs Opus 4.8 at 83.4%, human expert baseline at 72.4%) - Terminal-Bench 2.1: 80.4% (a 20.7-point jump over Sonnet 4.6) - Humanity's Last Exam with tools: 57.4% (vs Opus 4.8 at 57.9% — essentially tied) - GDPval-AA v2 (knowledge work): Sonnet 5 slightly edges ahead of Opus 4.8 — a first for a mid-tier model
Pricing that reshapes the competitive landscape:
Through August 31, introductory pricing is $2 per million input tokens and $10 per million output tokens — significantly undercutting Opus 4.8, OpenAI's GPT-5.5 ($15 output), and Google's Gemini 3.1 Pro. After that, standard pricing of $3/$15 applies, matching Sonnet 4.6's rate, though the new tokenizer produces 1.0–1.35× more tokens from the same text.
Three breaking changes developers must know before migrating:
1. Adaptive thinking is always on — extended thinking is now default, changing response format and latency profiles 2. Temperature and sampling parameters have been removed — any code setting these will error 3. New tokenizer — per-request budget enforcement must be recalibrated
Production reliability that matches the hype:
Cursor co-founder Sualeh Asif reported that Sonnet 5 "stays on plan, follows conventions, and ships clean multi-step changes at efficient cost." Zapier senior engineer Daniel Shepard described a two-part Salesforce automation that "used to stall halfway" now completing end-to-end.
On safety, Sonnet 5 shows lower rates of hallucination, sycophancy, and cooperation with misuse than its predecessor, while refusing malicious requests more reliably — critical for autonomous agentic workflows where human oversight is reduced.
The launch comes as the AI industry enters a defining quarter, with Anthropic targeting an October IPO and needing to demonstrate that frontier-adjacent agentic capability at sustainable pricing can drive enterprise adoption at scale.
