Model

Claude Opus 4.8

CLOSED

Anthropic · Claude 4 family · released May 28, 2026

Model id claude-opus-4-8. Introduced 'dynamic workflows' in Claude Code and a 'fast mode' (2.5x speed) priced at $10/$50 per MTok, three times cheaper than prior fast-mode pricing. Regular (non-fast) pricing unchanged from Opus 4.7 at $5/$25. Best-in-class computer-use/browser-agent model at launch (84% on Online-Mind2Web, self-reported).

ReasoningCodingVisionFunction callingTool useAgentic

2670.2

Elo · rank #24

Parameters

Undisclosed

Active params

—

Context

1M tokens

Architecture

Dense transformer, architecture undisclosed; adaptive thinking controlled by a runtime effort parameter (low/medium/high/xhigh/max)

License

Proprietary

Languages

—

API price (in/out)

$5 / $25

Modalities

text · vision

Benchmark results

Bar shows position within the tracked field; marker = field best

AA-BriefcaseAgents1346#6

best: Claude Opus 5 · 1720

Agents' Last ExamAgents45.2%#8

best: GPT-6 Astra · 59.3%

Aider PolyglotCoding79.1%#7

best: Claude Opus 4.5 · 89.4%

ARC-AGI-3Reasoning1.5%#7

best: GPT-6 Astra · 62.7%

Arena EloHuman preference1481#17

best: Claude Fable 5 · 1504

AutomationBenchAgents17.0%#18

best: DeepSeek-V4.1-Flash · 54.8%

BrowseCompAgents84.3%#11

best: Kimi K3 · 91.2%

DeepSWECoding59.0%#21

best: Gemini 4 Argon · 77.9%

ExploitBenchAgents40.0%#7

best: GPT-6 Astra · 100.0%

FrontierBenchAgents21.1%#4

best: Claude Opus 5 · 43.3%

FrontierCodeCoding46.5%#9

best: Claude Opus 5.5 · 54.6%

FrontierMath (Tiers 1-3)Math47.2%#9

best: GPT-5.6 Sol · 89.0%

FrontierMath (Tier 4)Math31.3%#7

best: GPT-5.6 Sol · 83.0%

GDPval-AAAgents1890#2

best: Claude Fable 5 · 1932

GPQA DiamondReasoning93.6%#7

best: GPT-6 Astra · 96.0%

Humanity's Last ExamReasoning49.8%#12

best: Claude Opus 5.5 · 67.7%

LiveCodeBenchCoding87.8%#16

best: DeepSeek-V4-Pro (Think Max) · 93.5%

MMMUVision86.6%#3

best: Claude Fable 5 · 89.3%

OSWorld 2.0Agents55.7%#10

best: Claude Opus 5.5 · 81.8%

OSWorld-VerifiedAgents83.4%#3

best: Claude Fable 5 · 85.0%

SWE-bench MultilingualCoding84.4%#5

best: Claude Opus 5.5 · 93.9%

SWE-bench MultimodalCoding38.4%#6

best: Claude Opus 5.5 · 61.4%

SWE-bench ProCoding69.2%#5

best: Claude Opus 5.5 · 89.9%

SWE-bench VerifiedCoding88.6%#11

best: Claude Opus 5 · 97.0%

Terminal-Bench 2.0Coding74.6%#26

best: DeepSeek-V4.1-Flash · 90.6%

Terminal-Bench 3.0Agents21.1%#6

best: GPT-5.6 Sol · 34.6%

Toolathlon (Verified)Agents76.2%#4

best: GLM-5.3-Flash · 78.4%

USAMOMath96.7%#1

best: this model · 96.7%

Run it locally

Closed weights — available via API only. No local deployment.

Input / M tok

$5

Output / M tok

$25

Claude 4 family

Elo progression across releases

Claude Opus 4May 20251637.6 Claude Sonnet 4May 20251533.4 Claude Opus 4.1Aug 20251700.2 Claude Sonnet 4.5Sep 20251905.6 Claude Haiku 4.5Oct 20251518.6 Claude Opus 4.5Nov 20252187.9 Claude Opus 4.5 (High)Nov 20252220.5 Claude Opus 4.5 (Medium)Nov 20252266.8 Claude Opus 4.6Feb 20262420.5 Claude Sonnet 4.6Feb 20262370.1 Claude Opus 4.7Apr 20262650.0 Claude Opus 4.8May 20262670.2

Resources

DOCShttps://platform.claude.com/docs/en/about-claude/models/overview APIAvailable via Anthropic API

API price $5/$25 · each benchmark row carries its own source badge (see methodology)