Claude Sonnet 5 — AI Model Review & Pricing
Mid-tier agentic coding model from Anthropic with a 1,000,000 token (~977K) context window, $2.00/M input pricing, and 76.1% on Novel Compiler evaluation. Balances speed and intelligence for professional tasks, agentic search, and code generation, trailing Opus-class models on capability benchmarks but offering substantial cost savings.
Specifications
| Attribute | Value |
|---|---|
| Lab | Anthropic |
| Tags | Intelligent Reasoning Coding Agentic Image Input Long Context |
| Release Date | 2026-06 |
| Context Window | 1,000,000 tokens |
| Input Price / 1M | $2.00 |
| Output Price / 1M | $10.00 |
| Input Modalities | Text, Image, File |
| Output Modalities | Text |
Strengths
- 76.1% on Novel Compiler evaluation (vendor-reported)
- 1,000,000 token (~977K) context window with 128k output
- $2.00/M input tokens — 85% cheaper than GPT-5.5 Pro
- 96.65% harmless response rate on API (vendor-reported)
- Up to 90% cost savings with prompt caching
Weaknesses
- Trails Opus 4.8 and Mythos 5 on capability benchmarks
- Higher over-refusal rate than frontier models
- New tokenizer reduces effective text capacity per context
- Priority Tier not available
Best For
- Agentic coding and software development
- Professional knowledge work at scale
- Cost-sensitive enterprise deployments
- Long-context retrieval and analysis
In Depth: Claude Sonnet 5
Benchmark Performance
Claude Sonnet 5 earns a 9.4/10 LMRank (#6), scoring 76.1% on Novel Compiler evaluation — ahead of Opus 4.7 (70.4%) but behind Mythos 5 (85.3%). It trails GPT-5.5 Pro (rank #5, 9.5/10) on overall capability benchmarks while improving over Sonnet 4.6 across coding, agentic search, and multimodal reasoning.
Claude Sonnet 5 achieves 76.1% on the Novel Compiler evaluation, surpassing Opus 4.7's 70.4% but trailing Mythos 5's 85.3% (vendor-reported). On Anthropic's internal coding and agentic search evals, it shows clear gains over Sonnet 4.6 while remaining below Opus 4.8 and Mythos-class models across most capability benchmarks. Its harmless response rate reaches 96.65% API and 99.20% on Claude.ai.
Pricing & Value
Claude Sonnet 5 costs $2.00 per million input tokens and $10.00 per million output tokens through August 31, 2026. This introductory pricing is roughly 85% cheaper than GPT-5.5 Pro's $30.00 in / $180.00 out, while delivering a 9.4/10 LMRank score.
Comparing price-per-LMRank-point, Claude Sonnet 5 at $2.00 in / $10.00 out per M tokens costs $0.21 per point versus GPT-5.5 Pro at $30.00 in / $180.00 out per M tokens ($3.16 per point). Prompt caching can reduce costs up to 90%, and batch processing offers 50% savings. Standard pricing after August 31, 2026 rises to $3.00 in / $15.00 out, still undercutting comparable peers.
Who Should Use This
Claude Sonnet 5 fits coding agents who need 128k output tokens and 1M context at low cost. Knowledge workers gain professional-task performance with batch processing savings. Teams prioritizing maximum capability over price should consider Opus 4.8 or Mythos 5 instead.
- Coding agents: Strong agentic coding and 128k output tokens for complex workflows; trails Opus 4.8 on hardest benchmarks.
- Knowledge workers: Reliable professional task performance with 1M context window; over-refusal rate higher than frontier models.
- Cost-conscious teams: Introductory pricing at $2.00/M input tokens; standard pricing rises after August 31, 2026.
- Not for: Cyber or high-stakes safety applications where Mythos 5 or Opus 4.8's lower refusal rates are needed.
Release & Version History
Claude Sonnet 5 arrives June 30, 2026 as Anthropic's mid-tier option, succeeding Sonnet 4.6. It fills the gap between premium Opus models and value-focused tiers, offering a 1,000,000 token (~977K) context window at prices 87% lower than Opus 4.8.
Claude Sonnet 5, released June 30, 2026, succeeds Sonnet 4.6 as Anthropic's mid-tier model. It introduces a new tokenizer producing ~30% more tokens for the same text and supports 128k maximum output tokens. The model sits below Opus 4.8 (LMRank 9.7, #2) and Opus 4.7 (9.6, #3) in the Anthropic lineup, with no evidence of a 'Sonnet 5.5' variant as of search date. The official name is 'Claude Sonnet 5,' not 'Claude 5 Sonnet.'