Released 2026-05-28 by Anthropic

Claude Opus 4.8
Sharper Agentic Judgement · 4x Fewer Code Flaws

Sharper, more reliable judgement on agentic tasks, 4x fewer code flaws than Opus 4.7, Online-Mind2Web 84%, native 1M context — at the same price as Opus 4.7 ($5/$25 per MTok)

#Opus 4.8 #claude-opus-4-8 #agentic coding #strong multimodal #long-horizon tasks #migration must-read

Key Highlights

+13%
4x Fewer Code Flaws

4x less likely than Opus 4.7 to allow flaws in code — a step-change in production code reliability

64.3%
Online-Mind2Web 84%

84% task success on computer-use / browser-agent workloads, with sharper, more reliable agentic judgement

3.75 MP
Legal Agent Benchmark

Highest score recorded and the first model to break 10% overall on the all-pass standard

$5 / $25
Same Price

$5 input / $25 output per MTok — identical to Opus 4.7, no price increase

Coding Leap: Reliability You Can Ship

From long-horizon autonomy to complex tool calls, Opus 4.8 makes agentic code work dramatically more dependable — far fewer defects to catch downstream

4x Fewer Code Flaws vs 4.7

Opus 4.8 is 4x less likely than Opus 4.7 to allow flaws in code, meaning fewer review cycles and fewer regressions reaching production

CursorBench Across Every Effort Level

Exceeds all prior Opus models on CursorBench across every effort level — more headroom whether you optimize for latency or depth

Sharper Agentic Judgement

Sharper and more reliable when performing agentic tasks: better decisions about when to act, when to verify, and when to ask

Super-Agent: End-to-End Completion

On the Super-Agent benchmark, Opus 4.8 is the only model to complete every case end-to-end without dropping the task

Online-Mind2Web 84%

84% success on computer-use / browser-agent tasks — reads dense UI, navigates, and acts reliably across multi-step flows

⭐ Improved Honesty

Improved honesty means fewer confident-but-wrong claims about its own work — it flags uncertainty instead of papering over it

Strong Multimodal & Long-Context Handling

Opus 4.8 pairs strong multimodal understanding with better long-context handling across native 1M context — send images directly via API, no parameter switch needed

Computer-use Agents Reading Dense Screenshots

Strong multimodal grounding lets agents read UI detail and act reliably, contributing to the 84% Online-Mind2Web result

Long-Context Stability

Better long-context handling across native 1M context — large repos, long transcripts, and multi-document tasks stay coherent

Document & Chart Understanding

Reads charts, tables, and document layouts as part of multimodal reasoning, extracting structure and detail in one pass

Legal Agent Benchmark Leadership

Highest score recorded and first to break 10% on the all-pass standard — evidence of reliable judgement on demanding domain tasks

Platform Capabilities

xhigh

Effort Tier: xhigh

The xhigh tier between high and max offers a finer-grained reasoning-depth vs latency trade-off. Carried forward into Opus 4.8

/ultrareview

/ultrareview Deep Code Review

Claude Code command for an independent review session that runs through changes end-to-end, finding bugs and design issues

task_budgets

Adaptive Thinking

Adaptive thinking lets Claude self-allocate reasoning depth across long tasks instead of a fixed token budget

auto_mode

Fast Mode Default

Opus 4.8 is now the Claude Code Fast Mode default (replacing 4.7), bringing sharper judgement to everyday fast iterations

Migration Guide (⭐ Key)

Upgrading from Opus 4.7 to Opus 4.8 is a drop-in replacement (change model ID to claude-opus-4-8) — here is what to keep in mind

1. Drop-in Model ID Swap

Change the model id from claude-opus-4-7 to claude-opus-4-8 — no other config change required

2. Expect Sharper Judgement

Opus 4.8 brings sharper agentic judgement and improved honesty; re-validate prompts and harnesses to take advantage of the more reliable decisions

3. Use Adaptive Thinking

Use thinking={type:"adaptive"} with the effort parameter; the legacy thinking={type:"enabled", budget_tokens:N} form remains deprecated

4. 1M Context & Fast Mode

Native 1M context carries over; Opus 4.8 is now the Fast Mode default. A 1M-context variant id (claude-opus-4-8[1m]) is also available

Code Example: Model ID Migration
❌ Deprecated
client.messages.create(
    model="claude-opus-4-7",
    thinking={"type": "enabled", "budget_tokens": 10000}
)
✅ Recommended
client.messages.create(
    model="claude-opus-4-8",
    thinking={"type": "adaptive"},
    effort="xhigh"   # available since 4.7
)

vs GPT-5.4 / Gemini 3.1 Pro

Same-tier flagship comparison (Anthropic has not published a full SWE-bench Pro figure for the newest Opus — the headline gains are agentic reliability and 4x fewer code flaws)

Metric Opus 4.8 GPT-5.4 Gemini 3.1 Pro
Code reliability 4x fewer flaws vs 4.7 See OpenAI See Google
Input $ / MTok $5 See OpenAI See Google
Output $ / MTok $25 See OpenAI See Google
Context window Native 1M 272K / 1M beta 1M
Computer-use (Online-Mind2Web) 84%

Get Opus 4.8 via QCode.cc

Stable developer platform with official pricing, ready to use

Same Price $5/$25

QCode.cc bills at Anthropic's official rates with no multiplier markup

Full Support for New Parameters

Full pass-through of xhigh effort, adaptive thinking, and other Opus 4.8 parameters

Drop-in Switch 4.7 to 4.8

Change model ID from claude-opus-4-7 to claude-opus-4-8, no other config changes needed

China-Direct with Failover

Multi-node smart routing + circuit breakers, avoiding instability of direct official API access from within China

Try Opus 4.8 Now

Sign up for QCode.cc to get a stable Claude Opus 4.8 developer platform