Anthropic
Most UsedClaude 4.7 — natural prose, multi-file engineering, and the new Computer Use loop.
Anthropic's Claude 4.7 family is the industry benchmark for naturalistic prose, long-form writing, and agentic engineering. Sonnet 4.6 ranks #1 on the LMSYS writing leaderboard at 1,643 Elo and reaches roughly 98% of Opus 4.7's capability at one-fifth the price — the practical sweet spot for most enterprise agents. The 2026 'Computer Use' feature lets Claude operate web browsers and software UIs directly, unlocking a class of agents that were previously impossible.
Models
Claude Opus 4.7
LatestLeads SWE-bench Verified for autonomous engineering
Context Window
1M
Max Output
32K tokens
Input Price
$5.00 / 1M
Output Price
$25.00 / 1M
Input Types
Text, Images, PDFs
Output Types
Text, Code, JSON
Claude Sonnet 4.6
Recommended1,643 Elo on writing — 98% of Opus at 1/5 the price
Context Window
1M
Max Output
16K tokens
Input Price
$3.00 / 1M
Output Price
$15.00 / 1M
Input Types
Text, Images
Output Types
Text, Code
Claude Haiku 4.5
Lowest-latency instruction following
Context Window
200K
Max Output
8K tokens
Input Price
$0.25 / 1M
Output Price
$1.25 / 1M
Input Types
Text, Images
Output Types
Text, Code
Use Cases
Brand-Voice Writing
Sonnet 4.6 is the benchmark for natural prose — long-form blog posts, marketing copy, and editorial work that doesn't read like AI.
Multi-File Engineering
Opus 4.7 leads SWE-bench Verified — give it a repo and a ticket, get back a coherent multi-file change set.
Computer Use Agents
Claude can drive a browser or desktop app directly — extracting data, filing tickets, or operating legacy SaaS without a custom integration.
Long-Document Reasoning
1M token context lets Opus and Sonnet ingest entire contracts, research corpora, or codebases in one pass.
Why use Anthropic on Nagent?
Nagent adds enterprise orchestration, observability, and workflow automation on top of Anthropic's raw model capabilities.
All three Claude tiers exposed — pick the right cost / quality tradeoff per agent task
Native tool-use and structured JSON output are first-class in our agent runtime
Automatic prompt caching cuts repeated-context cost by up to 90%
Computer Use loop wired into our browser-agent skill — no extra integration work
How to access Anthropic on Nagent
Open Agent Studio
Navigate to Agent Studio in your Nagent workspace.
Select Anthropic
Choose Anthropic under the Model Configuration panel and pick your Claude tier.
Configure & Deploy
Set temperature, max tokens, and system prompt — then wire your agent to any tool or workflow.
Common questions about Anthropic
Real buyer and developer questions, answered. Click any item to expand.
Which Claude tier should I use for my agent?
Sonnet 4.6 is the default for most production work — it hits roughly 98% of Opus 4.7's capability at one-fifth the price. Pick Opus 4.7 when the task is autonomous multi-file engineering or where complexity meaningfully rewards the price step. Pick Haiku 4.5 for high-volume routing, classification, or low-stakes summarisation.
What is Claude's Computer Use feature, and is it production-ready?
It lets Claude operate a browser or desktop app directly — clicking, typing, reading the screen — without a custom integration per app. Excellent for legacy SaaS automation. Accuracy still varies on complex flows, so wire it through Nagent's human-in-the-loop approval queue for high-stakes work.
How does prompt caching reduce my Anthropic costs?
Repeated context (system prompts, tool schemas, large reference documents) is cached for 5 minutes. Cache reads cost roughly 10% of cache writes. For typical agent loops where the same context recurs every turn, total spend drops by 60–80% with caching enabled.
What's the rate limit for Anthropic models on Nagent?
Nagent pools across multiple Anthropic API keys and tiers, so typical workloads see no rate limiting. During burst spikes the router auto-fails over to OpenAI / Mistral as configured fallbacks — your agents keep responding.
Ready to use Anthropic inside your agents?
Get started in minutes — no API key management required.