Claude Opus 4.8
Claude Opus 4.8 is an Anthropic model in the Opus tier of the Claude family, accepting text and image input with a 1,000,000-token context window and tool calling support.
| Provider | Input / 1M | Output / 1M | Cached / 1M |
|---|---|---|---|
| $2.29 | $11.47 | $0.229 | |
| $2.50 | $12.50 | $0.250 | |
| $5.00 | $25.00 | - | |
| $5.00 | $25.00 | - | |
| $5.00 | $25.00 | $0.500 | |
| $5.00 | $25.00 | $0.500 | |
| $5.00 | $25.00 | $0.500 |
Prices updated daily. Last check: Oct 23, 2026
Compare API pricing for every Claude model →Claude Opus 4.8 pricing by provider
Input, output, and batch rates, plus alternatives, for one provider at a time.
Performance & Benchmarks
Source: Artificial Analysis →Reasoning & Knowledge
- GPQA Diamond92.0%
- Humanity's Last Exam48.7%
Coding
- SciCode54.4%
Agentic & Tool Use
- Terminal-Bench Hard58.3%
- Terminal-Bench v2.184.6%
- τ²-bench94.4%
- τ-bench Banking34.2%
Instruction & Long Context
- IFBench62.2%
- Long-Context Reasoning77.7%
Benchmarks measured Oct 2026. Scores are independent evaluations, not vendor-reported.
Model Details
General
- Creator
- Anthropic
- Family
- Opus
- Tier
- Flagship
- Context Window
- 1.0M
- Knowledge Cutoff
- Jan 2026
- Modalities
- Text, Image
Capabilities
- Tool Calling
- Yes
- Open Source
- No
- Aliases
- claude-opus-4-8, Claude Opus 4.8, anthropic.claude-opus-4-8
Strengths & Limitations
Strengths
- 1,000,000-token context window, five times the 200K window of earlier Claude Opus generations
- Accepts image input in addition to text, supporting screenshots, scanned documents, charts, and diagrams
- Tool calling is supported, enabling function-calling loops and agent orchestration
- Opus-tier positioning targets complex reasoning and multi-step coding rather than high-volume throughput
- Knowledge cutoff recorded as January 2026, more recent than earlier Claude 4.x releases
- Available through multiple hosting routes, including Anthropic's API and cloud marketplace aliases such as anthropic.claude-opus-4-8
- Very large context reduces the need for aggressive chunking or retrieval preprocessing on long documents and codebases
Limitations
- Opus-tier models are typically the most expensive option within the Claude family — check the pricing table on this page
- Throughput and time-to-first-token are not populated in our benchmark data, so speed cannot be compared directly here
- Filling a large share of the 1M-token window increases per-request cost and generally raises latency
- Model weights are not distributed for self-hosting; access is through hosted APIs
- Knowledge cutoff of January 2026 means later events must be supplied via context or tools
Key Features
About Claude Opus 4.8
Common Use Cases
Claude Opus 4.8 fits workloads where the input is large and the reasoning chain is long: reviewing or refactoring code across many files, analyzing contracts, filings, or research corpora that run to hundreds of thousands of tokens, and driving agents that call tools repeatedly before producing a final answer. Image input makes it usable for mixed-media tasks such as reading scanned pages, interpreting charts inside reports, or working from UI screenshots during front-end debugging. Because Opus-tier models generally cost more per token than Sonnet or Haiku, a common pattern is tiered routing — a smaller Claude model handles classification, extraction, and routine chat, while Opus 4.8 is reserved for the requests that need the full million-token window or the hardest multi-step reasoning.
Frequently Asked Questions
How much does Claude Opus 4.8 cost?
Pricing varies by provider and by pricing type — input tokens, output tokens, cached reads, and batch processing are typically billed at different rates, and long-context requests consume many more tokens per call. See the pricing table on this page for current rates across the providers we track.
What is Claude Opus 4.8 best used for?
It suits long-context and multi-step work: whole-repository code analysis and refactoring, review of large document sets, and agentic pipelines that chain many tool calls. The 1,000,000-token context window and image input make it a fit for tasks that would otherwise require heavy chunking or a separate OCR step.
How does Claude Opus 4.8 differ from earlier Claude Opus models?
The most concrete difference in our data is context length: Claude Opus 4.8 records a 1,000,000-token window, versus the 200K windows associated with earlier Opus 4.x releases. Its knowledge cutoff is also later, recorded as January 2026.
Does Claude Opus 4.8 support images and tool calling?
Yes. Our metadata lists text and image input modalities, and tool calling is supported, so the model can be used in function-calling and agent frameworks that pass images alongside text prompts.
Should I use Claude Opus 4.8 or a Sonnet-tier Claude model?
Opus-tier models are positioned for the hardest reasoning and longest-context tasks and generally carry higher per-token costs; Sonnet-tier models are the usual choice for latency- and volume-sensitive traffic. Many deployments route selectively, sending only requests that need the full context window or deep multi-step reasoning to Opus 4.8.
How fast is Claude Opus 4.8?
We do not currently have populated throughput or time-to-first-token figures for this model, so we cannot state output speed here. If latency matters, compare provider-reported numbers for the specific endpoint you plan to use, and note that large context fills typically increase time to first token.