Anthropic Launches Low-Cost Agent Model Claude Sonnet 5
모델 출시 | Wed Jul 01 2026 00:00:00 GMT+0000 (Coordinated Universal Time) | 2 sources
Anthropic unveiled an agent-specialized mid-size model that delivers performance close to Opus 4.8 at a significantly lower price.
Analysis
[Anthropic] unveiled Claude Sonnet 5 [1][2]
- The most agentic Sonnet model to date
- Supports tool use across browsers and terminals
- and autonomous execution
- Delivers performance approaching Opus 4.8
[Claude Sonnet 5] significantly improved agent performance benchmarks [1][2]
- Scored 63.2% on agentic coding benchmark
- Improved over Sonnet 4.6 (58.1%)
- approaching Opus 4.8 (69.2%)
- Slightly surpassed Opus 4.8 on knowledge work benchmark
- Comprehensive improvements over Sonnet 4.6 on BrowseComp and OSWorld-Verified
[Anthropic API] introduced low-cost promotional pricing [1][2]
- $2 per 1M input tokens
- $10 per 1M output tokens
- Introductory pricing applies until August 31
- 2026
- Will increase to $3 input
- $15 output afterward
- Cheaper than GPT-5.5 and Gemini 3.1 Pro
[Claude Platform] deployed Sonnet 5 as the default model across all plans [1]
- Set as the default model for Free and Pro plans
- Available to Max
- Team
- and Enterprise users
- Accessible via Claude Code and Claude Platform
- API access via claude-sonnet-5
[Competitive Landscape] triggered full-scale price competition on agent capabilities [2]
- OpenAI unveiled GPT-5.6 Sol preview
- Google Gemini 3.5 Flash pivoted to agents
- Agent capabilities have become baseline expectations at every price point
- Differentiation is shifting to low cost and reliability
[Safety Evaluation] improved safety in agentic environments [1]
- Reduced rate of undesirable behaviors compared to Sonnet 4.6
- Generally safer in agentic contexts
- Cybersecurity task performance significantly lower than Opus
- Detailed evaluations published in the System Card