---
title: "Claude Opus 4.8"
type: model
id: "claude-opus-4.8"
provider: "Anthropic"
model_type: "proprietary"
api_model_id: "claude-opus-4-8"
release_date: "2026-05-28"
description: "Anthropic's most capable Opus-tier model for complex reasoning and long-horizon agentic coding. 69.2% SWE-bench Pro, 1M context on by default, user-level effort control, adaptive thinking."
last_updated: "2026-06-10"
last_verified: "2026-06-10"
knowledge_cutoff: "2026-01"
availability_status: "available"
deprecated: false
tool_schema_format: "anthropic"
pricing_confidence: "high"
model_listing_confidence: "high"
benchmark_confidence: "low"
context_window: "1M tokens"
website: "https://anthropic.com"
license: "Proprietary"
modality:
- "text"
- "image"
tags:
- "anthropic"
- "proprietary"
- "text"
- "image"
pricing:
  input: "$5.00 / 1M tokens"
  output: "$25.00 / 1M tokens"
  input_per_1m: 5
  output_per_1m: 25
  currency: "USD"
  cache_read_per_1m: 0.5
  batch_input_per_1m: 2.5
  batch_output_per_1m: 12.5
  note: "Same rate card as Opus 4.5 through 4.7. Long-context input pricing applies over 200K tokens. Fast Mode (2.5x speed) is $10/$50. US-only inference is 1.1x. Minimum cacheable prompt is 1,024 tokens."
benchmarks:
  reasoning: 97
  coding: 98
  math: 94
  writing: 96
  multilingual: 89
  speed: 65
capabilities:
- "function_calling"
- "vision"
- "web_search"
- "code_execution"
- "structured_output"
- "streaming"
- "prompt_caching"
- "tool_search"
- "long_context"
- "reasoning"
- "adaptive_thinking"
sources:
- title: "Anthropic Claude models overview"
  url: "https://platform.claude.com/docs/en/about-claude/models/overview"
- title: "Anthropic API pricing"
  url: "https://platform.claude.com/docs/en/about-claude/pricing"
- title: "Claude Opus 4.8"
  url: "https://www.anthropic.com/claude/opus"
benchmark_sources:
- title: "Anthropic: Claude Fable 5 and Claude Mythos 5 (comparison table)"
  url: "https://www.anthropic.com/news/claude-fable-5-mythos-5"
- title: "AI Future Ready benchmark methodology"
  url: "https://ai-future-ready.com/guides/benchmark-methodology"
best_for:
- "Complex coding projects"
- "Long-horizon agentic workflows"
- "High-stakes enterprise tasks"
- "Knowledge work"
---

# Claude Opus 4.8

The default frontier workhorse. Released May 28, 2026 -- just six weeks after Opus 4.7 -- Opus 4.8 is Anthropic's most capable Opus-tier model and the best price-to-capability option at the top of the market. At 69.2% SWE-bench Pro it leads every generally available model except Fable 5, and it keeps the $5/$25 rate card that has held since Opus 4.5.

The headline feature is user-level effort control: `low`, `high` (the default), `extra`, and `max`. Higher effort consumes more tokens at the same per-token rate, so you tune cost-versus-depth per request instead of switching models. Adaptive thinking adjusts reasoning depth to task complexity automatically. The 1M context window is on by default with 128K max output, and Fast Mode gives a 2.5x-speed variant at $10/$50 -- three times cheaper than Fast Mode was on Opus 4.7.

Migration math matters here. Coming from Opus 4.7, the upgrade is config-only: same API surface, same tokenizer, same pricing. Coming from Opus 4.6 or earlier, the Opus 4.7 tokenizer change still applies -- the same text produces roughly 30% more tokens, which raised effective costs by up to 35% for some workloads. Budget for that before assuming flat costs.

**When to pick something else:** Fable 5 is meaningfully stronger (80.3% vs 69.2% SWE-bench Pro) if you can justify double the price and a 30-day retention requirement. Sonnet 4.6 at $3/$15 remains the right call for the large majority of production traffic that doesn't need Opus-level depth. For rapid-fire low-stakes work, Haiku 4.5 at $1/$5 is the volume play.