Claude Opus 4.8

Opus 4.8 is the safe choice when your prompts, parsers and evals were built around its output and you do not want surprises. Anthropic released it in May 2026 as the step after Opus 4.7. Early testers, according to Anthropic, found it more dependable on agent tasks, and it cleaned up the overly wordy code comments and tool-calling problems that some people hit on 4.7.

Newer Opus releases exist now, but on Kadraris you can keep 4.8 in production and call the newer one with the same key for comparison.

What you can use it for

• Production jobs whose prompts were written and tested on Opus 4.8
• Coding agents that make many tool calls inside one session
• Reading a long report or contract bundle in a single 1M-token call
• A/B runs against Claude Opus 5.5 on identical inputs before you switch

Settings and limits

• Context window: 1,000,000 tokens; output up to 128K tokens.
• Thinking is adaptive and the effort setting starts at high.
• A fixed thinking token budget from older code is rejected, so remove it when you port requests.
• Anthropic allows system messages further down the conversation on this model, not only at the start.
• Works through /v1/chat/completions and /v1/messages, including Claude Code.

Example request

from openai import OpenAI

client = OpenAI(
    base_url="https://kadraris.com/v1",
    api_key="sk-kd-...",
)
resp = client.chat.completions.create(
    model="claude-opus-4.8",
    messages=[{"role": "user",
               "content": "Check this diff for off-by-one "
                          "errors and missing tests."}],
)
print(resp.choices[0].message.content)

Paste your own diff after the instruction. For CLI use, see the Claude Code guide.

Questions

Is Opus 4.8 cheaper than the newer Opus?

Not necessarily. Opus 4.8 runs at $3.78 / $18.90 per 1M tokens; check the rate of Claude Opus 5.5 on its own page, because a newer release does not always cost more. Failed calls are free either way.

How do I move to a newer Opus without breaking things?

Send the same set of real inputs to both models with one key and diff the outputs. Switch only when your checks pass. The Sonnet 5 vs Opus 4.8 page also helps if a cheaper model might be enough.

Why did my old thinking budget start failing?

Opus 4.8 uses adaptive thinking only. Drop the manual budget field and, if you need shallower reasoning, lower the effort level instead.

Claude Opus 4.8 in your code

from openai import OpenAI

client = OpenAI(base_url="https://kadraris.com/v1", api_key="sk-kd-...")
resp = client.chat.completions.create(
    model="claude-opus-4.8",
    messages=[{"role": "user", "content": "Suggest three opening shots for a short film about a cafe."}],
)
print(resp.choices[0].message.content)

curl https://kadraris.com/v1/chat/completions \
  -H "Authorization: Bearer $KADRARIS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "claude-opus-4.8", "messages": [{"role": "user", "content": "Suggest three opening shots for a short film about a cafe."}]}'

import OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://kadraris.com/v1", apiKey: process.env.KADRARIS_API_KEY });
const resp = await client.chat.completions.create({
  model: "claude-opus-4.8",
  messages: [{ role: "user", content: "Suggest three opening shots for a short film about a cafe." }],
});
console.log(resp.choices[0].message.content);

export ANTHROPIC_BASE_URL=https://kadraris.com
export ANTHROPIC_AUTH_TOKEN=sk-kd-...
export ANTHROPIC_MODEL=claude-opus-4.8
claude