Claude Opus 5 is Anthropic's flagship for the hardest work: architecture, multi-file refactors, long agentic runs. Anthropic released it on 24 July 2026, and codexkey had it the same day under the same id, claude-opus-5, with no queue and no separate whitelist.
What Claude Opus 5 is
Opus 5 replaced Opus 4.8 as the default pick for heavy work. At codexkey you call it by id claude-opus-5.
What changed
Three changes show up right away.
- Speed. On long agentic runs the gap from the previous revision is visible without a benchmark, especially with many steps.
- Tools on the fly. The tool set used to be fixed at the start of a conversation. Now you can change it mid-dialogue without a new session.
- Automatic fallback to Opus 4.8. If the safety classifier flags a request, the call does not fail but quietly goes to Opus 4.8. Budget and logic should assume part of the traffic is served by the neighboring model.
When Opus 5 earns its price
Use the flagship where many relationships must stay in mind:
- reading an unfamiliar repository to grasp its architecture
- a refactor touching a dozen files coherently
- debugging races and intermittent bugs with no obvious cause
- agent planning, where the model picks its own next steps
- long reviews of code or contracts where context must survive
If a task fits "take this input and convert it by pattern", Opus 5 is overkill.
How to call Opus 5
One base URL: https://api.codexkey.online. Both formats work, Anthropic Messages and OpenAI Chat Completions, with the same sk_cdx_... key.
Via the Messages API:
curl https://api.codexkey.online/v1/messages \
-H "x-api-key: sk_cdx_..." \
-H "anthropic-version: 2023-06-01" \
-H "content-type: application/json" \
-d '{
"model": "claude-opus-5",
"max_tokens": 1024,
"messages": [{"role": "user", "content": "Review this repository structure and propose a refactor plan"}]
}'
Via the OpenAI SDK in Python:
from openai import OpenAI
client = OpenAI(api_key="sk_cdx_...", base_url="https://api.codexkey.online/v1")
resp = client.chat.completions.create(
model="claude-opus-5",
messages=[{"role": "user", "content": "Review this repository structure"}],
)
print(resp.choices[0].message.content)
Via the Anthropic SDK in TypeScript:
import Anthropic from "@anthropic-ai/sdk";
const client = new Anthropic({
apiKey: "sk_cdx_...",
baseURL: "https://api.codexkey.online",
});
const msg = await client.messages.create({
model: "claude-opus-5",
max_tokens: 1024,
messages: [{ role: "user", content: "Plan a refactor of the auth module" }],
});
One key in all three examples; only the client and the URL path change.
What it costs
Directly from Anthropic, Opus 5 costs $5 per million input tokens and $25 per million output tokens. At codexkey the same model runs $0.3 per million input and $0.3 per million output. That is several times cheaper, so we call it our actual price, not an introductory discount.
The site calculator estimates a bill for your own load: pick the model, enter volumes, read the sum. Current rates for every model are on the codexkey pricing page; article figures go stale, that page updates itself. The dashboard splits usage by key, so separate keys show separate spend.
Honest note: when a cheaper model is enough
Even at our price a flagship is not free. For sorting inbound messages, converting formats, parsing logs and boilerplate from a pattern, reach for claude-sonnet-5 or gpt-5.6-luna: the quality gap barely shows and the bill drops noticeably. A workable scheme: Opus 5 reads the task once and writes a plan, a cheaper model does the grunt work.
FAQ
What is Claude Opus 5?
Anthropic's flagship Claude model, released on 24 July 2026. It took over from Opus 4.8 as the first pick for hard tasks. At codexkey you call it by id claude-opus-5.
How is Opus 5 different from Opus 4.8?
It is faster on long agentic runs, lets you change tools mid-conversation and quietly switches to Opus 4.8 instead of returning an error when the safety classifier fires.
How much does Claude Opus 5 cost?
Directly from Anthropic, at the time of writing, $5 per million input tokens and $25 per million output tokens. At codexkey it costs less; exact rates are on the pricing page.
When should I use Opus 5 and when a cheaper model?
Use Opus 5 to keep a whole system in view: an unfamiliar repository, a refactor across a dozen files, intermittent bugs, agent planning. For classification, format conversion and boilerplate, claude-sonnet-5 or gpt-5.6-luna is enough.
How do I call Opus 5 through codexkey?
Set the base URL to https://api.codexkey.online and the model to claude-opus-5. Both API formats take the same sk_cdx_... key. Examples are above, in "How to call Opus 5".
In short
Opus 5, out since 24 July 2026, is faster than the previous revision, changes tools mid-conversation and falls back to Opus 4.8 when the safety classifier fires. Use it for whole-system work and hand routine tasks to cheaper models.