Overview
Claude Sonnet 4.8 is a mid-range model released by Anthropic on **2026/5/28**, skipping version 4.7 and directly named 4.8—Anthropic officially explains that "4.8 has a generational leap in Agent capabilities compared to 4.6, deserving a larger version number jump." This version is a key product for Claude to rebuild market trust after the **Fable 5 incident (around 4/24 when Claude Opus 4.7 faced widespread benchmark cheating controversy)**—Sonnet 4.8 focuses on "affordable, stable, solid coding ability," clearly not pursuing Opus-level highest intelligence, but firmly capturing the "daily workhorse" and "budget-sensitive" markets.
**⚠️ Important Note**: **On 2026/6/30, Anthropic released [Claude Sonnet 5](/tools/claude-sonnet-5.html)** (performance close to Opus 4.8, limited-time pricing $2/M input + $10/M output, promotional period until 8/31). **Sonnet 4.8 has been replaced by Sonnet 5 as the recommended model**, but 4.8 is still available in the API (model ID `claude-sonnet-4-8`) for existing integrations. **New projects are advised to directly choose Sonnet 5**—but note that on Sonnet 5's launch day, there were negative reviews about "some Agent tasks being overtaken by Qwen 3.7-Max and MiniMax M3," so it's safer to test and compare before deciding.
**Key capabilities of Sonnet 4.8**:
- **Primary coding model positioning**: SWE-bench Verified approximately 55-58% (figures from public benchmarks), about 6-8 percentage points improvement over Sonnet 4.6, top tier among mid-range models
- **Significantly enhanced Agent capabilities**: This is the core reason for the version jump from 4.7 to 4.8, with notable improvements in tool call accuracy, multi-step reasoning stability, and long-context task retention
- **Outstanding cost-effectiveness**: Mid-range pricing + stable output makes it one of the default backend options for AI coding tools like Cursor, Windsurf, Codex, and Cline
If you are a developer who has already integrated Sonnet 4.8 into your workflow and is cost-sensitive, you can continue using it; if you are a new user or pursuing the latest capabilities, go directly to Sonnet 5.
Key Features
- Version naming skip (4.6→4.8): Anthropic explicitly states it's due to a generational leap in Agent capabilities, not a marketing gimmick
- Primary coding model: SWE-bench Verified approximately 55-58%, 6-8 percentage points improvement over Sonnet 4.6
- Significantly enhanced Agent capabilities: Comprehensive improvements in tool call accuracy, multi-step reasoning stability, and long-context retention
- Outstanding cost-effectiveness: Mid-range pricing, one of the default backends for mainstream AI coding tools like Cursor/Windsurf/Codex/Cline
- Replaced by Sonnet 5: On 2026/7/1, Anthropic released Sonnet 5, performance close to Opus 4.8; 4.8 is still available in the API but no longer the recommended model
Use Cases
- Existing production projects that have already integrated Sonnet 4.8 and do not plan immediate migration
- Budget-sensitive coding Agent scenarios seeking cost-effectiveness
- Enterprise customers requiring high model stability and not wanting the latest version
- Default backend for AI coding tools (Cursor/Cline/Windsurf, etc.)
Pros
- Solid coding ability, top tier among mid-range models
- Generational leap in Agent capabilities compared to 4.6
- High cost-effectiveness, widely integrated into mainstream AI coding tools
- Stable output style, low hallucination rate, production-friendly
- Trust-rebuilding product from Anthropic after the Fable 5 incident
Pricing
API pricing: Input $3/M tokens, Output $15/M tokens (historical prices, subject to Anthropic official website). **After Sonnet 5's release (7/1), Sonnet 4.8 can still be called via the `claude-sonnet-4-8` model ID, but Anthropic officially recommends new projects directly choose Sonnet 5** (limited-time $2/M input + $10/M output, promotional period until 8/31, then price increase). Claude web version and app subscribers can use it via model switching.
Summary
Sonnet 4.8 is a key product for the Claude family to rebuild market trust after the Fable 5 incident, focusing on "affordable, stable, solid coding." **But note it has been replaced by Sonnet 5 as of 2026/7/1**—new projects are advised to directly choose Sonnet 5 (stronger capabilities, cheaper limited-time price); existing integrations that are not in a hurry to migrate can continue using 4.8. If you see the "Sonnet 4.8" option in tools like Cursor, Windsurf, or Cline, it's a good choice, just be aware that the newer Sonnet 5 is available.
Version History
- Sonnet 5 released, 4.8 relegated to secondary (2026-07-01): Anthropic releases Sonnet 5 (performance close to Opus 4.8, limited-time price $2/M+$10/M until 8/31); Sonnet 4.8 still available in API but no longer the recommended model
- Claude Sonnet 4.8 released (2026-05-28): Skipped 4.7 directly named 4.8; core product for Anthropic's trust rebuilding after Fable 5 incident; generational leap in Agent capabilities, solid coding ability, outstanding cost-effectiveness