Major Model Deprecations and Migration Guide

Date: 2026-09-04 / (1405-06-13)

Summary

AvalAI is retiring a broad set of deprecated models across OpenAI, Google, Anthropic, Stability AI, and other providers as of September 4, 2026. This notice lists the affected model IDs and the recommended replacements, including the GPT-5.6 family for OpenAI chat, the Gemini "Nano Banana" image family for Imagen 4, and the new audio models for TTS and transcription.

---

Details

OpenAI, Google, Anthropic, and several other providers have scheduled or completed the retirement of a large group of models. AvalAI is aligning its catalog with these upstream changes, so requests to the deprecated IDs listed below will fail. This notice groups the most impactful deprecations and gives you a concrete replacement and migration path for each.

For the complete list of affected models, always refer to the Deprecated Models page, which is the authoritative, continuously updated source.

OpenAI GPT chat models → GPT-5.6 family

The GPT-5 -chat tags and earlier GPT-5.x chat snapshots are no longer available. Requests using gpt-5-chat, gpt-5-chat-latest, gpt-5.1-chat, gpt-5.2-chat, or gpt-5.3-chat will fail. Use the base model names without -chat, and migrate to the GPT-5.6 family for new work.

Deprecated model	Use instead
gpt-5-chat / gpt-5-chat-latest	gpt-5.6-sol
gpt-5.1-chat	gpt-5.6-terra
gpt-5.2-chat / gpt-5.3-chat	gpt-5.6-sol
gpt-4o, gpt-4o-mini (legacy)	gpt-5.6-terra / gpt-5.6-luna

The GPT-5.6 family is the current recommended default for general chat and reasoning:

• gpt-5.6-sol — highest-capability model for difficult coding, expert analysis, and agentic workflows (pro mode via reasoning.mode: pro)
• gpt-5.6-terra — balanced production model for everyday reasoning and knowledge work
• gpt-5.6-luna — cost-efficient model for high-volume tasks

Bash

curl https://api.avalai.ir/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "gpt-5.6-sol",
    "messages": [
      {
        "role": "user",
        "content": "Your prompt here."
      }
    ]
  }'

Python

from openai import OpenAI

client = OpenAI(api_key="your-avalai-api-key", base_url="https://api.avalai.ir/v1")

completion = client.chat.completions.create(
    model="gpt-5.6-sol",  # replaces gpt-5-chat / gpt-5.x-chat
    messages=[
        {"role": "user", "content": "Your prompt here."},
    ],
)

print(completion.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 completion = await client.chat.completions.create({
  model: "gpt-5.6-sol", // replaces gpt-5-chat / gpt-5.x-chat
  messages: [{ role: "user", content: "Your prompt here." }],
});

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

Google Imagen 4 image models → Nano Banana (Gemini image) family

Google has deprecated the Imagen 4 generation models. Google recommends the Nano Banana (Gemini native image) family as the replacement.

Deprecated model	Use instead
imagen-4.0-generate-001	gemini-3.1-flash-image (Nano Banana 2)
imagen-4.0-ultra-generate-001	gemini-3-pro-image (Nano Banana Pro)
imagen-4.0-fast-generate-001	gemini-3.1-flash-image (Nano Banana 2)
imagen-4.0-fast-generate-preview-06-06	gemini-3.1-flash-lite-image (Nano Banana 2 Lite)

See the full guide: Generate and Edit Images with the Nano Banana Series.

OpenAI TTS and transcription models → new audio models

OpenAI has scheduled the retirement of its legacy TTS and transcription models. The deprecated tts-1, tts-1-hd, gpt-4o-mini-tts, whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe, and gpt-4o-transcribe-diarize are replaced by the new audio and transcription models.

Deprecated model	Use instead
gpt-4o-mini-tts, tts-1, tts-1-hd	gpt-audio-1.5 (or Gemini TTS below)
gpt-4o-transcribe, gpt-4o-mini-transcribe	gpt-transcribe
gpt-4o-transcribe-diarize	gpt-live-transcribe
whisper-1	gpt-transcribe / gpt-live-transcribe

For conversational voice applications, see Building Conversational Apps with Audio Models, which uses gpt-audio-1.5, gpt-audio, and gpt-audio-mini with the /v1/audio/speech → /v1/responses → /v1/audio/transcriptions pipeline. Gemini TTS (gemini-2.5-pro-tts, gemini-2.5-flash-tts) is also available when your workflow standardizes on Gemini audio.

Anthropic Claude 4 models

The Claude 4 generation is retired. Migrate to the current Claude models.

Deprecated model	Use instead
claude-opus-4, claude-opus-4-1, anthropic.claude-opus-4-20250514-v1:0	claude-opus-4-7
claude-sonnet-4, anthropic.claude-sonnet-4-20250514-v1:0	claude-sonnet-4-6

Stability AI image models → Gemini / GPT Image

The Stability AI image generation and editing services (stability.sd3-large-v1:0, stability.stable-image-*, stability.sd3-5-large-v1:0, stability.stable-image-ultra-v1:1, stability.stable-image-core-v1:1) are retired.

Deprecated model	Use instead
stability.sd3-5-large-v1:0, stability.stable-image-ultra-v1:1	gemini-3.1-flash-image or gpt-image-2
stability.stable-image-core-v1:1, other stability.stable-image-*	gemini-3.1-flash-image

Other provider deprecations

A wide range of additional models are retired. The table below lists the most common ones; see Deprecated Models for the complete catalog.

Provider	Deprecated	Use instead
DeepSeek	deepseek.r1-v1:0, deepseek-r1-0528, deepseek-v3-0324	deepseek-v4-flash / deepseek-v4-pro
Groq	groq.playai-tts, groq.playai-tts-arabic	gpt-audio-1.5 or Gemini TTS
Mistral	codestral-2501, mistral-ocr-2503, mistral-ocr-2505	mistral-ocr-4-0 / mistral-ocr-latest
Alibaba	qwq-32b, qwen3-235b-a22b-thinking-2507	qwen3.5-plus / qwen3-8-max
Moonshot	kimi-k2-0905	kimi-k3
Meta (Bedrock)	meta.llama2-*, meta.llama3-*	llama-4-scout-17b-16e-instruct
NVIDIA NIM	nvidia_nim.* (selected)	See /v1/models for current NIM routes
Cloudflare	cf.gemma-3-12b-it, cf.qwq-32b	See /v1/models for current Cloudflare routes

How to migrate

1. Identify usage: Review your code for any of the deprecated IDs above.
2. Pick the replacement: Use the tables above to choose the target model.
3. Update the model ID: Change only the model parameter; endpoints and request shape are unchanged.
4. Test: Verify quality, latency, and cost with your own prompts before production rollout.
5. Pin for reproducibility: Use the explicit model ID (not a -latest alias) when you need controlled rollouts.

---

Related Links

• Deprecated Models
• Generate and Edit Images with the Nano Banana Series
• Building Conversational Apps with Audio Models
• GPT-5.6 Family announcement
• Model catalog and pricing