Claude API / OPERATOR API
claude-haiku-4-5-20251001
Versioned prices, current public supply and measured request outcomes.
PUBLISHED TARIFF
Pay for recorded usage
- 5-minute cache write / million
- $1.25
- 1-hour cache write / million
- $2.00
- Spending hold per request
- $1.00
- Seller share of buyer charge
- 0%
- Price version
bdaabb79-1def-47b5-be99-60d94413bc40
The recorded price stays fixed for each admitted request. A hold reserves spending room; it is not an estimated charge. Earnings accrue in the ledger; a paid transfer is a separate event.
ROUTE & LIMITS
What this route supports
- Protocol
- Native Messages · JSON and SSE
- Request bounds
- 1 MB · 100 messages · 32 client tools
- Output setting
- 1–4,096 tokens
- Maximum provider context
- Not verified
- Tools and images
- Client tools, PNG, thinking and cache TTLs live-verified on the published Haiku source
API-funded capacity; no seller subscription discount. JPEG, WebP and GIF inputs are accepted but have not each passed a live test. JSON/schema output, server tools and continuation sessions are unavailable.
OBSERVED PERFORMANCE
Public traffic · last 24 hours
Includes acceptance tests and failed requests, across prices for this model. Private requests are excluded. This sample describes observed traffic; it is not an uptime guarantee.
| Measurement | Samples | 10th percentile | Median | 90th percentile |
|---|---|---|---|---|
| First text | 3 | 0.472 s | 0.630 s | 0.664 s |
| Recorded duration | 3 | 0.534 s | 0.690 s | 0.724 s |
First text measures dispatch to the first text event. Duration measures dispatch to the recorded finish, including recovery when relevant. Each row excludes missing measurements; percentiles interpolate the sample. 3 settled · 0 released. Outcomes and accounting states are separate counts.
PLAN YOUR USAGE
Estimate a request
Change the example token counts to match your expected workload. The calculation uses the same price version and exact arithmetic as billing.
ESTIMATED BUYER CHARGE
- Input / million
- $1.00
- Cached / million
- $0.10
- Output / million
- $5.00
- Spending hold / request
- $1.00
1 eligible public sources in this snapshot.
This is a planning estimate, not a quote, reservation or inference request. Actual provider usage determines the charge; unused holds are released. Subscription output limits are hints.
Price versionbdaabb79-1def-47b5-be99-60d94413bc40DEVELOPER SETUP
Connect your client
Choose a published model and protocol. Your application key stays in your terminal.
Endpoint https://aftercore.io/v1/messages
Set AFTERCORE_API_KEY to your Aftercore key. Use a public key for market supply; private keys can only use your own eligible seats. Automatic retries are disabled. Running this example purchases inference at the displayed model’s price.
// Each run purchases a new request unless you reuse AFTERCORE_REQUEST_ID. Inspect Usage before retrying.
import { randomUUID } from 'node:crypto';
import Anthropic from '@anthropic-ai/sdk';
const headers = { 'Idempotency-Key': process.env.AFTERCORE_REQUEST_ID || randomUUID(), 'X-Aftercore-Price-Version': "bdaabb79-1def-47b5-be99-60d94413bc40", ...(process.env.AFTERCORE_SESSION ? { 'X-Aftercore-Session': process.env.AFTERCORE_SESSION } : {}) };
const client = new Anthropic({ apiKey: process.env.AFTERCORE_API_KEY, baseURL: "https://aftercore.io", maxRetries: 0 });
const stream = await client.messages.create({
"model": "claude-haiku-4-5-20251001",
"max_tokens": 32,
"stream": true,
"messages": [
{
"role": "user",
"content": "Say hello in one short sentence."
}
]
}, { headers });
for await (const event of stream) {
if (event.type === 'content_block_delta' && event.delta.type === 'text_delta') process.stdout.write(event.delta.text);
}Download exampleOnly the supported streaming surface is shown. Output limits on subscription routes are hints; the versioned spending hold is the buyer debit ceiling. Inspect Usage after any interruption before sending another request.
Supply can change
Public keys use only eligible listed supply. Private keys use only the owner's eligible sources. Listing, quota and key policies are checked again at admission.
Billing follows the receipt
Failures before output release holds without earnings. Partial output is billed only with valid usage. Missing or conflicting usage is waived after five minutes; it never becomes an estimated settled charge.
Snapshot 2026-09-14T01:55:04.960Z · Reload to check current supply and measurements.