Claude Sonnet 4.5

Providers

Every routable endpoint for this model, with the rates the gateway bills against. One endpoint today, so there is no routing decision to make — when there is more than one, the router picks between them on price, health and latency.

Effective Pricing Computed

What input tokens actually cost once prompt caching is working. This is arithmetic over Anthropic's published prices, not a measurement — cached input bills at $0.30 per million instead of $3, so the blended rate falls as more of your prefix is reused: list × (1 − hit) + cache read × hit.

Performance Sample data

Throughput, round-trip latency and time to first token. Nothing here has been measured — there are no probes running and no traffic to learn from yet, so these are placeholders showing what the hydration layer will fill in.

• 112tokens / secondoutput, median
• 390mstime to first tokenmedian
• 390msp50 latencyend to end
• 1240msp99 latencyend to end

Uptime Sample data

Share of requests that completed without an upstream error, over a rolling 30-day window. Sample data — the gateway has not made enough requests to have an opinion.

• 99.93%success raterolling 30d
• 0.07%error raterolling 30d

Apps Sample data

How many distinct applications sent traffic to this model each week. We deliberately do not list app names: the per-app breakdown arrives with label passthrough on the event stream, and inventing customer names to fill the gap would be worse than an empty section.

Activity Sample data

Token volume and request traffic over the trailing four weeks, rolled up from the event stream. Sample data until the stream has something real in it.

This model ranks #9 across the catalog this week — see the full rankings.

Call it

Standard OpenAI client, one changed line. Streaming is the default path — non-streaming is just collecting the stream.

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.aigate.dev/v1",
  apiKey: process.env.AIGATE_API_KEY,
});

const stream = await client.chat.completions.create({
  model: "anthropic/claude-sonnet-4-5",
  messages: [{ role: "user", content: "Hello" }],
  stream: true,
});

for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? "");
}

Capabilities

• Streaming
• Tool use
• Vision
• PDF input
• Citations
• Extended thinking
• Prompt caching
• Batch API

Accepted parameters

• max_tokens
• stream
• stop_sequences
• system
• tools
• tool_choice
• thinking
• temperature
• top_p
• top_k

Specification

Model ID

anthropic/claude-sonnet-4-5

Upstream ID

claude-sonnet-4-5

Context window

200,000 tokens

Max output

64,000 tokens

API dialect

anthropic — translated for you

Knowledge cutoff

Jan 2025

Min cacheable prefix

1,024 tokens

Also available on

claude-api, bedrock, google-cloud

FAQ

Closest alternatives

By input price. Switch between any of them by changing the model string.

• Claude Sonnet 51M context$3 / $15
• Claude Sonnet 4.61M context$3 / $15
• Claude Opus 51M context$5 / $25