New Flagship Model Added: Claude Opus 5.5

Date: 2026-09-24 / (1405-07-02)

Summary

Anthropic's Claude Opus 5.5 is now available on AvalAI for long-running coding agents and knowledge work. It supports Chat Completions, Messages, and Responses, with always-on adaptive thinking, a 1M-token input window, and lower input, output, and cache-read prices than Opus 5.

Details

Anthropic: Claude Opus 5.5

Use claude-opus-5-5 for demanding software engineering, research, and professional analysis. Anthropic released the model on September 22, 2026; this announcement covers its availability on AvalAI on September 24.

• Long-running coding and agents: Designed for multi-step repository changes, migrations, audits, debugging, and verification before declaring a task complete.
• Knowledge work: Supports research, document analysis, technical writing, and professional deliverables. Check important claims against source material.
• Vision and tools: Accepts text and images, supports PDF input, function calling, structured output, and prompt caching, and returns text. Computer-use workflows need a compatible tool definition and an application-controlled execution environment; vision does not make this an image-generation model.
• Adaptive reasoning: Thinking is always on, with medium as the provider's default effort. Tune effort through the documented Anthropic controls rather than copying settings from another provider.
• Provider-reported efficiency: Anthropic reports around 40% lower cost on typical workloads at default settings and over 30% faster output generation than Opus 5. These are provider workload claims, not a guaranteed AvalAI latency improvement or a blanket discount on every request.

Specification	Value
AvalAI model ID	claude-opus-5-5
Input context window	1,000,000 tokens
Maximum output	128,000 tokens
Thinking	Always-on adaptive thinking
Default effort	medium
Provider knowledge and training cutoff	June 2026
Account access	Tier 1 or higher

The model has native 1M context; do not add an old context beta header merely to select that capacity. Input, conversation history, tool results, and output budgeting still need to fit the selected route's limits. See the Anthropic provider guide.

Supported endpoints

Endpoint	AvalAI support
v1/chat/completions	Full support
v1/messages	Full support
v1/responses	Full support

Use the same exact model ID on all three endpoints. OpenAI SDK clients use https://api.avalai.ir/v1; Anthropic SDK clients use https://api.avalai.ir. Endpoint support does not automatically enable every upstream hosted tool, beta header, Fast mode, or Batch feature. This release does not announce an automatic replacement of claude-opus-5 or claude-fable-5-1.

AvalAI pricing

All prices below are USD per 1 million tokens.

Model	Input	Cached input	Cache creation input	Output
claude-opus-5-5	$4.00	$0.20	$8.00	$20.00

Input and output rates are 20% lower than the documented Opus 5 rates, and cache reads are 60% cheaper. Cache creation is a separate charge: use AvalAI's $8.00 rate, not the upstream $5.00 five-minute cache-write rate. A matching price does not by itself establish a cache lifetime; follow the prompt caching guide for supported cache controls.

Upstream Fast mode pricing, Batch discounts, and the 300K-output Batch beta are not included in this AvalAI announcement. The documented output limit here remains 128,000 tokens. Check pricing and rate limits for current account details.

Migration notes

Moving from Opus 5 is not always a model-name-only change:

1. Thinking cannot be disabled. Use adaptive thinking and begin with medium effort. Do not send a disabled-thinking configuration or a fixed extended-thinking budget. Omit unsupported sampling fields such as temperature and top_p, and do not use assistant prefill.
2. Forced tool use returns an error. Remove configurations that require a tool call or force a named tool. Let the model select tools normally and validate tool arguments in your application.
3. Thinking blocks are bound to the model and conversation. Preserve the complete assistant content, thinking blocks, signatures, and tool-call context when continuing a conversation. Do not edit or transplant them into another model or conversation. Review preserved thinking before changing history handling.
4. The old computer tool is incompatible. Anthropic documents that computer_20251124 is not accepted on the Claude API and Google Cloud. Check the current tool schema and its support on your AvalAI route instead of carrying an old beta configuration forward.
5. Progress text has changed shape. Text between tool calls now arrives in thinking blocks whose text is empty at the default display setting. A progress UI must not treat that silence as a stalled request. Use tool-status events, and verify supported display controls if visible progress text is required.

Ask for concise conclusions, citations, and verification evidence—not hidden chain-of-thought. Keep human approval for destructive actions and treat external documents and tool results as untrusted inputs. Anthropic's upstream safeguard fallbacks and verification programs are separate from AvalAI account access; this announcement does not promise unrestricted upstream feature parity. See the reasoning guide and official migration notes.

API request and response examples

Keep AVALAI_API_KEY on the server. These minimal requests omit optional thinking overrides and use the model's default adaptive behavior. Allow enough output budget for both reasoning and the final answer.

Chat Completions

Bash

curl https://api.avalai.ir/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "claude-opus-5-5",
    "max_completion_tokens": 8192,
    "messages": [
      {"role": "user", "content": "Review a staged database migration and list verification and rollback checks."}
    ]
  }'

Python

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["AVALAI_API_KEY"],
    base_url="https://api.avalai.ir/v1",
)

response = client.chat.completions.create(
    model="claude-opus-5-5",
    max_completion_tokens=8192,
    messages=[
        {
            "role": "user",
            "content": "Review a staged database migration and list verification and rollback checks.",
        }
    ],
)

print(response.choices[0].message.content)

JavaScript

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.AVALAI_API_KEY,
  baseURL: "https://api.avalai.ir/v1",
});

const response = await client.chat.completions.create({
  model: "claude-opus-5-5",
  max_completion_tokens: 8192,
  messages: [
    { role: "user", content: "Review a staged database migration and list verification and rollback checks." },
  ],
});

console.log(response.choices[0].message.content);

Illustrative Chat Completions response

This is an illustrative example, not a captured live response. Token counts and the exchange rate are examples, not a current quote. The completion count includes reasoning tokens; the hidden reasoning itself is not shown. Real responses may include additional thinking or tool blocks that must be preserved in conversation history.

{
  "id": "chatcmpl-claude-opus-5-5-example",
  "created": 1790208000,
  "model": "claude-opus-5-5",
  "object": "chat.completion",
  "system_fingerprint": null,
  "choices": [
    {
      "finish_reason": "stop",
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Verify backup restoration, test schema compatibility, migrate a small batch, compare row counts and checksums, monitor errors and latency, and stop or roll back if predefined checks fail.",
        "annotations": []
      }
    }
  ],
  "usage": {
    "prompt_tokens": 100,
    "completion_tokens": 400,
    "total_tokens": 500,
    "completion_tokens_details": {
      "reasoning_tokens": 300
    },
    "prompt_tokens_details": {
      "cached_tokens": 0,
      "text_tokens": 100,
      "audio_tokens": null,
      "image_tokens": null
    }
  },
  "estimated_cost": {
    "unit": "0.0084000000",
    "irt": 840,
    "exchange_rate": 100000
  }
}

With no cache reads or writes, the example cost is (100 × $4 + 400 × $20) / 1,000,000 = $0.0084. Use the returned usage and cost-tracking guidance for actual billing.

Native Messages

The native API requires an output budget. Read text blocks by type rather than assuming the first content block is text; this display-only filtering must not replace the full content retained for future turns.

Bash

curl https://api.avalai.ir/v1/messages \
  -H "Content-Type: application/json" \
  -H "x-api-key: $AVALAI_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -d '{
    "model": "claude-opus-5-5",
    "max_tokens": 8192,
    "messages": [
      {"role": "user", "content": "Review a staged database migration and list verification and rollback checks."}
    ]
  }'

Python

import os
from anthropic import Anthropic

client = Anthropic(
    api_key=os.environ["AVALAI_API_KEY"],
    base_url="https://api.avalai.ir",
)

message = client.messages.create(
    model="claude-opus-5-5",
    max_tokens=8192,
    messages=[
        {
            "role": "user",
            "content": "Review a staged database migration and list verification and rollback checks.",
        }
    ],
)

print("".join(block.text for block in message.content if block.type == "text"))

JavaScript

import Anthropic from "@anthropic-ai/sdk";

const client = new Anthropic({
  apiKey: process.env.AVALAI_API_KEY,
  baseURL: "https://api.avalai.ir",
});

const message = await client.messages.create({
  model: "claude-opus-5-5",
  max_tokens: 8192,
  messages: [
    { role: "user", content: "Review a staged database migration and list verification and rollback checks." },
  ],
});

console.log(message.content.filter((block) => block.type === "text").map((block) => block.text).join(""));

Responses

For OpenAI-compatible Responses clients, use input instead of messages and read response.output_text. Do not copy OpenAI-specific reasoning fields or assume hosted-tool availability solely because the endpoint is supported.

Bash

curl https://api.avalai.ir/v1/responses \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "claude-opus-5-5",
    "max_output_tokens": 8192,
    "input": "Review a staged database migration and list verification and rollback checks."
  }'

Python

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["AVALAI_API_KEY"],
    base_url="https://api.avalai.ir/v1",
)

response = client.responses.create(
    model="claude-opus-5-5",
    max_output_tokens=8192,
    input="Review a staged database migration and list verification and rollback checks.",
)

print(response.output_text)

JavaScript

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.AVALAI_API_KEY,
  baseURL: "https://api.avalai.ir/v1",
});

const response = await client.responses.create({
  model: "claude-opus-5-5",
  max_output_tokens: 8192,
  input: "Review a staged database migration and list verification and rollback checks.",
});

console.log(response.output_text);

Related resources

• Claude Opus 5.5 model reference
• Anthropic provider guide
• Chat Completions API, Messages API, and Responses API
• Reasoning guide and prompt caching
• AvalAI pricing and rate limits
• Anthropic system card