Official commercial channel
Claude Opus 5 API
A million tokens that never degrade, five-level effort control, zero data retention.
Context
1M
Max output
128K
Serving precision
Vendor-managed
Licence
Proprietary
Claude Opus 5 was released on 24 July 2026 and carries the strongest overall benchmark profile Anthropic has reported for an Opus model: a 1-million-token context window, up to 128,000 tokens of synchronous output and a new five-level effort control. It scores 100% on needle-in-a-haystack — most long-window models quietly lose reliability past 200K, and this one does not. Vendor list pricing is $5 per million input and $25 per million output, half of Fable 5, and it supports zero data retention.
Proprietary, provided through the vendor’s official commercial licensing. Contract, invoice, audit trail — not an account pool and not a reverse-engineered client protocol.
Specifications
- Model ID
- claude-opus-5
- Published by
- Anthropic
- How it is served
- Official commercial channel
- Licence
- Proprietary
- Context
- 1M
- Max output
- 128K
- Serving precision
- Vendor-managed
- Input modalities
- Text · Image
- Released
- 2026-07-24
- Price
- See pricing
What it is good at
- A million tokens you can actually use
- 100% on needle-in-a-haystack means a fact buried anywhere in a 1M-token document still comes back. That is a different thing from "1M advertised, leaky past 200K" — on long-document work it is the only number worth reading.
- Five effort levels, so cost is yours to set
- Pick reasoning depth per task instead of paying for the deepest chain on all traffic. Combined with routing, this is what makes Opus-class capability affordable in production.
- Zero data retention is supported
- This one matters to us: nothing is written to disk on our side, and Opus 5 supports zero retention upstream too. No segment of the path leaves you guessing where your conversations went.
Where it fits
- Repository-scale comprehension and automated refactoring
- Retrieval and deep analysis across million-token document sets
- Long-running agents that must respect strict constraints
- Workloads with hard data-retention compliance requirements
Three steps
Change one line of base_url, put the model ID below in the `model` field, and leave the rest of your code alone.
model: "claude-opus-5"
curl https://api.nexaix.net/v1/chat/completions \
-H "Authorization: Bearer $NEXAIX_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "claude-opus-5",
"stream": true,
"messages": [
{ "role": "user", "content": "Summarise this contract." }
]
}'import os
from openai import OpenAI
client = OpenAI(
# the only new line
base_url="https://api.nexaix.net/v1",
api_key=os.environ["NEXAIX_API_KEY"],
)
# everything below is your existing code
stream = client.chat.completions.create(
model="claude-opus-5",
stream=True,
messages=[
{"role": "user", "content": "Summarise this contract."}
],
)
for chunk in stream:
print(chunk.choices[0].delta.content or "", end="")import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://api.nexaix.net/v1", // the only new line
apiKey: process.env.NEXAIX_API_KEY,
});
// everything below is your existing code
const stream = await client.chat.completions.create({
model: "claude-opus-5",
stream: true,
messages: [
{ role: "user", content: "Summarise this contract." }
],
});
for await (const chunk of stream) {
process.stdout.write(chunk.choices[0]?.delta?.content ?? "");
}About this model
Is your Claude an official channel or an account pool?
An official commercial channel. This is the whole basis of the service: we do not mass-register consumer accounts and we do not reverse-engineer official client protocols. If you find a Claude relay priced far below official rates, it is almost certainly one of those two.
Do you avoid logging proprietary models too?
On our side, yes — nothing is written to disk, and we retain only the request ID, token counts, timestamp and status code needed for billing and debugging. Upstream depends on the vendor: Opus 5 supports zero data retention, which is uncommon among proprietary flagships and part of why we prefer recommending it.
Opus 5 or Fable 5?
Opus 5 lists at half Fable 5’s price, lands within 0.5% of Fable 5’s peak on CursorBench at half the cost per task, and supports zero data retention where Fable 5 currently carries a 30-day retention requirement. Unless you specifically need Fable 5’s ceiling, Opus 5 is the more sensible default.
Also worth a look
- Official commercial channelClaude Fable 5Anthropic’s tier above Opus — the one you pick when the ceiling is the point.claude-fable-5
- Official commercial channelGPT-5.6 SolThe frontier tier of the GPT-5.6 family: 1.05M context, built for long-horizon agents.gpt-5.6-sol
- Official commercial channelGPT-5.6 TerraThe everyday tier: the same window as Sol, at half the list price.gpt-5.6-terra
Run your own eval on it
Take the trial credit, run the evals you already have against this model, and then decide whether it gets your production traffic.
