Claude Haiku 4.5 brings Sonnet-class reliability to the cost-and-latency tier. 200K context, full vision input, native tool use and extended thinking on demand. Designed for high-volume agents, real-time chat, classification, simple coding subagents, and consumer applications where p50 latency matters more than peak reasoning.
Anthropic was founded in 2021 by Dario and Daniela Amodei. The Claude model family launched in March 2023 with Haiku as the lightweight tier optimized for low latency and cost. Successive Haiku generations have steadily closed the gap to Sonnet: Haiku 3 (March 2024), Haiku 3.5 (October 2024), Haiku 4 (May 2025) and Haiku 4.5 (late 2025/early 2026). Anthropic is structured as a Public Benefit Corporation focused on AI safety, with total funding exceeding $20 billion and a 2026 valuation above $150 billion.
Claude Haiku 4.5 is the fastest and cheapest model in the Claude 4.x family. Architecturally it is a smaller decoder-only Transformer distilled from the Sonnet/Opus 4 training stack, optimized aggressively for inference latency and throughput while retaining the full Claude feature set (vision, extended thinking, tool use). It was trained on the same multi-trillion-token curated corpus as its larger siblings with a knowledge cutoff in early 2026. Post-training combined RLHF, Constitutional AI, RLAIF and distillation from larger teacher models. Despite its small size, Haiku 4.5 supports the same 200K context window and tool-use API as Sonnet 4.6 and Opus 4.7, making it a drop-in cheaper option for high-concurrency workloads.
Paramètres
Undisclosed (estimated tens of billions of parameters)
Pretrained on the same curated multi-trillion-token mixture used by Sonnet and Opus 4.x. Post-training includes distillation from larger Anthropic models alongside RLHF, Constitutional AI and RLAIF. Knowledge cutoff in early 2026.
Licence: Proprietary commercial license via Anthropic API, Amazon Bedrock and Google Vertex AI.
curl https://railwail.com/api/v1/chat/completions \
-H "Authorization: Bearer $RAILWAIL_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "claude-haiku-4-5",
"messages": [
{
"role": "user",
"content": "Explain what a vector database is in two sentences."
}
],
"max_tokens": 1024
}'
import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["RAILWAIL_API_KEY"],
base_url="https://railwail.com/api/v1",
)
completion = client.chat.completions.create(
model="claude-haiku-4-5",
messages=[
{
"role": "user",
"content": "Explain what a vector database is in two sentences.",
},
],
max_tokens=1024,
)
print(completion.choices[0].message.content)
import OpenAI from "openai";
const client = new OpenAI({
apiKey: process.env.RAILWAIL_API_KEY,
baseURL: "https://railwail.com/api/v1",
});
const completion = await client.chat.completions.create({
model: "claude-haiku-4-5",
messages: [
{
role: "user",
content: "Explain what a vector database is in two sentences."
}
],
max_tokens: 1024
});
console.log(completion.choices[0].message.content);
// npm install railwail
import railwail from "railwail";
const rw = railwail(process.env.RAILWAIL_API_KEY);
const res = await rw.chat("claude-haiku-4-5", [
{ role: "user", content: "Explain what a vector database is in two sentences." },
], { max_tokens: 1024 });
console.log(res.choices[0].message.content);
