01Overview02Models03Rankings04Compare05Quiz06Guide

Model

Claude Opus 4.8 NO.02

State-of-the-art autonomous coding reliability at half the price of Fable 5.

CompareBack to the index

Anthropic · Claude · Frontier · Closed weights

Specification

Fig.01 — Claude — Anthropic

USA

Context window
1Mtokens
Max output
128Ktokens
Input price
$5/ 1M
Output price
$25/ 1M
Cached input
$0.50/ 1M
Throughput
—
Class
Frontier
Modalities
Text, Image, Pdf
Weights
Closed
Knowledge cutoff
Jan 2026
Released
May 2026

Overview

Claude Opus 4.8 is Anthropic's most capable Opus release and a state-of-the-art model for autonomous long-horizon agentic execution and coding reliability. It pairs frontier-grade planning and tool use with the durability to run extended agent loops without corrupting state, and adds parallel-subagent workflows and long-horizon memory. At half the price of Fable 5 it is the default frontier choice for demanding production coding agents.

Strengths

  • State-of-the-art autonomous coding reliability
  • Frontier long-horizon agentic execution
  • Strong price-to-capability at the frontier
  • Robust tool use and self-correction
  • Parallel subagent workflows and 1M context

Trade-offs

  • Pricier than balanced-tier models for everyday tasks
  • Not the fastest option on simple jobs
  • No audio/video input
  • Fable 5 edges it out on the very hardest tasks

Fit

Best for

  • Production coding agents and Claude Code workflows
  • Long-horizon autonomous workflows
  • Complex debugging and refactoring
  • Computer-use and browser automation
  • Agentic pipelines needing reliability

Not ideal for

  • Ultra-low-cost bulk workloads
  • Realtime low-latency chat

Capabilities

Ten axes, normalized 0–100 and scored the same way across the whole catalog — so a 78 here means what a 78 means anywhere else on the bench.

Fig.02 — Capability radar

0 — 100

Reasoning
96
Coding
96
Math
92
Writing
93
Knowledge
92
Speed
55
Agentic
96
Vision
88
Multilingual
90
Long Context
95

Benchmarks

Public results, with independent third-party runs marked. Bars normalize percentages against 100 and Elo ratings against a 1500 ceiling.

Independent index

Artificial Analysis Intelligence Index

Composite of ~9–10 independent evals · Artificial Analysis

56/ 100

BenchmarkResult
Artificial Analysis Intelligence IndexIndependentGeneral · Artificial Analysis
56%
SWE-bench VerifiedIndependentCoding · SWE-bench
88.6%
SWE-bench ProIndependentCoding · SWE-bench Pro
69.2%
τ-benchIndependentAgentic · τ-bench
74.6%
LiveBenchIndependentGeneral · LiveBench
77.2%
GPQA DiamondIndependentReasoning · Independent aggregators
93.6%
Terminal-Bench HardIndependentAgentic · Artificial Analysis
58.3%
LMArena EloIndependentGeneral · LMArena
1512Elo

Alternatives

Models in roughly the same class — the ones worth weighing against this record.

Compare side by side
01Claude Fable 5AnthropicAnthropic's most capable model — the reliability frontier for the hardest agentic work.02GPT-5.6 SolOpenAIOpenAI's GPT-5.6 flagship — its best coding-and-agentic model yet, for the hardest work.03Claude Opus 4.7AnthropicPrevious-generation frontier Opus, still highly reliable for agentic coding.

Index

PreviousClaude Fable 5NextClaude Opus 4.7
All models/EDUCATION