Claude Opus 4.8

Name: Claude Opus 4.8
Author: Anthropic

Released

2026-05-28

Last refreshed

2026-07-26

Status

Researched 109d ago

ProprietaryCommercial use: conditionalMultimodalCodingRAGAgentsLong contextVisionJSON / Tool useHighlight

Anthropic releases · 16 in the last 12 months · this family litChangelog →

Specifications

Family: Claude 4.8
Released: 2026-05-28
Context: 1M
Max output: 128K
Architecture: Decoder Only
Knowledge cutoff: 2026-01
Specialization: general
Openness: Proprietary
License: ProprietaryCommercial use: conditional
Weights: Not released
Code: Not released
Training: Fine-tuned

Created by

Anthropic

Developing safe and ethical AI systems.

San Francisco, California, United States

Founded 2021

Website

Pricing

Output / 1M

$25.00

Input / 1M

$5.00

Cheapest of 6 routes · Anthropic · cache read $0.500

Providers(6)

Anthropic AWS Bedrock GCP Vertex AI Microsoft Foundry OpenRouter Vercel AI Gateway

View 6 provider routes

Links

Website

About

Claude Opus 4.8 is Anthropic's flagship Claude 4.8 model, released May 28, 2026 for agentic coding, long-horizon reasoning, computer use, and professional knowledge work. It supports text and image inputs, adaptive reasoning, tool use, structured outputs, computer-use tools, prompt caching, Batch API, Dynamic Workflows parallel subagents, a 1M-token context window on Anthropic API/Bedrock/Vertex, and 128K max output. Reported benchmark scores: SWE-bench Pro 69.2%, SWE-bench Verified 88.6%, Terminal-Bench 2.1 74.6%, HLE with tools 57.9%, OSWorld-Verified 83.4%, GDPval-AA 1890 Elo, and MCP-Atlas 82.2%. Standard Anthropic API pricing is $5/M input and $25/M output.

Provider price ladder

Compare all 6

Compare API pricing across 4 providers for input and output tokens, batch, and cached reads when available.

Provider	Input / 1M	Output / 1M	Batch in / out	Cache	Route
Anthropic	$5.00	$25.00	$2.50 / $12.50	read $0.500 / 5m $6.25 / 1h $10.00	Serverless
OpenRouter	$5.00	$25.00	-	-	Serverless
AWS Bedrock	-	-	-	-	ServerlessPartial
GCP Vertex AI	-	-	-	-	ServerlessPartial

Available via routers & gateways(16)

LiteLLM

Gateway

Open-source Python SDK and proxy server that unifies 100+ LLM APIs behind a single OpenAI-compatible interface, with load balancing, cost tracking, and configurable failover.

Free OSSAnthropicGCP Vertex AIMicrosoft Foundry

OpenRouter

Hybrid

Unified hybrid gateway to 400+ models from 60+ providers via a single OpenAI-compatible API, with optional auto-routing that selects the best model per prompt.

PassthroughAnthropicGCP Vertex AI

Portkey

Gateway

Production AI gateway routing to 1,600+ LLMs with failover, load balancing, semantic caching, and guardrails; Apache 2.0 core is fully self-hostable with the complete feature set.

SubscriptionAnthropicGCP Vertex AIMicrosoft Foundry

AIRouter

Router

Commercial LLM router that analyzes incoming requests and routes to the optimal model for cost/quality/latency via a drop-in OpenAI-compatible API, with a privacy-preserving embedding mode that avoids sending prompt content.

Passthrough + feeAnthropicGCP Vertex AI

Amazon Bedrock Intelligent Prompt Routing

Router

AWS Bedrock's native intelligent prompt router that routes prompts between Anthropic Claude model tiers (Haiku/Sonnet) based on predicted task complexity, with no extra per-routing charge.

PassthroughAWS Bedrock

Azure AI Foundry Model Router

Router

Microsoft Azure AI Foundry's native model router that uses a trained ML model to route each prompt in real time to the optimal Azure-hosted model, with Balanced/Cost/Quality mode selection and automatic failover.

PassthroughMicrosoft Foundry

Capabilities

VisionMultimodalReasoningJSON / Tool useStructured OutputsCode ExecutionPrompt CachingBatch API

Benchmark peer barsfor Coding

SWE-bench ProRank 3 of 49

Claude Fable 5

80.3

Claude Opus 5

79.2

Claude Opus 4.8current

69.2

Qwen3.8-Max

67.7

Grok 4.5

64.7

SWE-bench VerifiedRank 4 of 92

Claude Fable 5

96.0

Claude Opus 5

96.0

Claude Mythos Preview

93.9

Claude Opus 4.8current

88.6

Claude Opus 4.7

87.6

LiveCodeBenchRank 9 of 68

DeepSeek V4 Pro

93.5

Gemini 3.1 Pro Preview

91.7

DeepSeek V4 Flash

91.6

Qwen3.7-Max

91.6

Claude Opus 4.8current

88.8

Benchmark scores(21)

Scores are benchmark-specific and are direction-aware: the same numeric gap can mean very different outcomes across suites. Use the leaderboard context and this model's provider route to decide whether the winning margin is meaningful for your workload.

Benchmark	Score	Version	Evaluation	Source
Finance Agent v2	53.9	—Observed 2026-05-28	—	Source
SWE-bench Verified	88.6	SWE-bench VerifiedObserved 2026-05-28	—	Source
SWE-bench Pro	69.2	SWE-bench ProObserved 2026-05-28	—	Source
Terminal-Bench 2.1	74.6	Terminal-Bench 2.1Observed 2026-05-28	—	Source
Google-Proof Q&A	93.6	GPQA DiamondObserved 2026-05-28	—	Source
Humanity's Last Exam — No Tools	49.8	HLE no toolsObserved 2026-05-28	—	Source
Humanity's Last Exam — With Tools	57.9	HLE with toolsObserved 2026-05-28	—	Source
BrowseComp — Single Agent	84.3	BrowseComp single-agentObserved 2026-05-28	—	Source
BrowseComp — Multi-Agent	88.5	BrowseComp multi-agent Dynamic WorkflowsObserved 2026-05-28	—	Source
OSWorld-Verified	83.4	OSWorld-VerifiedObserved 2026-05-28	—	Source
GDPval-AA	1890.0	GDPval-AA ELOObserved 2026-05-28	—	Source
MCP-Atlas	82.2	MCP-AtlasObserved 2026-05-28	—	Source
ARC-AGI-2 — High Effort	72.1	ARC-AGI-2 high effortObserved 2026-05-28	—	Source
LiveCodeBench	88.8	LiveCodeBenchObserved 2026-05-28	—	Source
CursorBench	63.8	CursorBench 3.1Observed 2026-06-30	Configuration: Opus 4.8 Max Harness: CursorBench 3.1 Evaluator: Cursor Confidence: confirmed Notes: Highest CursorBench 3.1 score across Cursor's published effort configurations for this base model.	Source
GeneBench-Pro	16.0	maxObserved 2026-06-30	—	Source
CursorBench	62.3	CursorBench 3.2Observed 2026-07-18	Configuration: Opus 4.8 Max Harness: CursorBench 3.2 productized Cursor-agent workflow Evaluator: Cursor Confidence: confirmed Cost/task: $5.77 Tokens/task: 71,411 Steps/task: 44 Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.	Source
CursorBench	59.4	CursorBench 3.2Observed 2026-07-18	Configuration: Opus 4.8 Extra High Harness: CursorBench 3.2 productized Cursor-agent workflow Evaluator: Cursor Confidence: confirmed Cost/task: $4.50 Tokens/task: 51,121 Steps/task: 40 Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.	Source
CursorBench	58.0	CursorBench 3.2Observed 2026-07-18	Configuration: Opus 4.8 High Harness: CursorBench 3.2 productized Cursor-agent workflow Evaluator: Cursor Confidence: confirmed Cost/task: $3.15 Tokens/task: 33,548 Steps/task: 33 Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.	Source
CursorBench	56.1	CursorBench 3.2Observed 2026-07-18	Configuration: Opus 4.8 Medium Harness: CursorBench 3.2 productized Cursor-agent workflow Evaluator: Cursor Confidence: confirmed Cost/task: $2.81 Tokens/task: 28,384 Steps/task: 32 Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.	Source
CursorBench	53.1	CursorBench 3.2Observed 2026-07-18	Configuration: Opus 4.8 Low Harness: CursorBench 3.2 productized Cursor-agent workflow Evaluator: Cursor Confidence: confirmed Cost/task: $2.02 Tokens/task: 19,624 Steps/task: 27 Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.	Source