New Models: GPT Image 2.5, DeepSeek V4.1 Flash, and Grok 4.6
Date: 2026-09-11 / (1405-06-20)
Summary
AvalAI now offers GPT Image 2.5 Flare and Sunburst, DeepSeek V4.1 Flash, and Grok 4.6 for image creation, multimodal reasoning, and agent workflows. DeepSeek off-peak pricing remains fixed around the clock. On September 14, requests to deepseek-v4-pro will route to deepseek-v4.1-flash; use the new model now.
Models and endpoint support
| Provider | AvalAI model ID | Supported endpoints |
|---|---|---|
| OpenAI | gpt-image-2.5-flare | v1/images/generations, v1/images/edits |
| OpenAI | gpt-image-2.5-sunburst | v1/images/generations, v1/images/edits |
| DeepSeek | deepseek-v4.1-flash | v1/chat/completions, v1/messages, v1/responses (partial support) |
| X.AI | grok-4.6 | v1/chat/completions, v1/messages, v1/responses (partial support) |
Partial Responses support is not full feature parity. Validate your specific tool, output-format, and conversation-continuation workflow before switching endpoints. Use the Images API to select either GPT Image 2.5 model directly; these image model IDs are not text models for Chat Completions or Responses.
OpenAI: GPT Image 2.5 Flare and Sunburst
- Flare is the default choice for general image generation, rapid prototyping, product imagery, and high-volume creative workflows. OpenAI reports up to 50% lower generation latency than GPT Image 2; actual AvalAI latency depends on the request and service conditions.
- Sunburst is suited to detailed creative work and premium editing workflows where tighter control matters more than generation time.
- Both improve natural lighting, textures, reference-subject preservation, targeted editing, and consistency across repeated edits. Send the previous image as an input when continuing an editing workflow.
ChatGPT-specific features such as Sketch and templates are not AvalAI API features. See the OpenAI provider documentation and image-generation guide.
DeepSeek: V4.1 Flash
deepseek-v4.1-flash supports native visual understanding, thinking and non-thinking modes, tool calls, JSON output, and prompt caching. Its 552B-parameter mixture-of-experts architecture uses a causal encoder–decoder design with 8B active parameters for input processing and 16B for output generation.
The AvalAI catalog lists 1,000,000 maximum input tokens and 393,216 maximum output tokens (384K in DeepSeek's terminology). Size requests to the selected route's limits. See the DeepSeek provider documentation.
X.AI: Grok 4.6
grok-4.6 is designed for long-running agents, engineering across large codebases, and knowledge work involving research, analysis, and refinement. It supports vision, reasoning, tool calling, structured output, and prompt caching. Interactive and visual work here means building applications and interfaces, not image generation.
The AvalAI catalog lists 500,000 maximum input tokens and 500,000 maximum output tokens. Do not carry over Grok 4.5's documented 1M input window when migrating. See the X.AI provider documentation.
Pricing
Text and reasoning models
Prices are in USD per 1 million tokens.
| Model / input length | Input | Cached input | Output |
|---|---|---|---|
deepseek-v4.1-flash — all hours | $0.15 | $0.003 | $0.60 |
grok-4.6 — up to and including 200K input tokens | $2.00 | $0.50 | $6.00 |
grok-4.6 — above 200K input tokens | $4.00 | $1.00 | $12.50 |
AvalAI offers DeepSeek's off-peak rates at every hour. The listed V4.1 Flash rates are already the lower rates: no peak-hour surcharge, scheduling requirement, or further halving applies. This continues our fixed off-peak pricing policy. Grok's higher-context prices are not a separate Fast variant.
GPT Image 2.5
Both models have the following token rates in the AvalAI catalog, in USD per 1 million tokens:
| Model | Text input | Cached text input | Image input | Cached image input | Text output | Image output |
|---|---|---|---|---|---|---|
gpt-image-2.5-flare | $5.00 | $1.25 | $8.00 | $2.00 | $0.00 | $30.00 |
gpt-image-2.5-sunburst | $5.00 | $1.25 | $8.00 | $2.00 | $0.00 | $30.00 |
The following estimates cover one output image at 1024x1024, not the complete request or a flat editing fee:
| Quality | Flare | Sunburst |
|---|---|---|
low | $0.00588 | $0.00588 |
medium | $0.01317 | $0.01317 |
high | $0.05268 | $0.05268 |
xhigh | $0.09366 | $0.09366 |
max | $0.21072 | $0.21072 |
Prompt tokens and any reference-image input tokens add to the output cost. Quality and dimensions affect output usage. A zero text-output rate does not mean image generation is free. See AvalAI pricing for the current catalog.
September 14: V4-Pro routing change
Starting September 14, 2026 at 04:00 UTC (1405-06-23), AvalAI will route deepseek-v4-pro requests to deepseek-v4.1-flash and bill them at V4.1 Flash rates. This is an upcoming change, not an already-completed migration as of this announcement.
Use deepseek-v4.1-flash now rather than starting new integrations on V4-Pro. Existing V4-Pro requests will be redirected automatically, but the underlying model changes; evaluate prompts, tool loops, output formats, latency, token budgets, and costs before the cutover. Keeping the old identifier will not pin the old V4-Pro model. The earlier V4-Pro specifications and prices describe the period before this change.
DeepSeek's upstream announcement also retires V4-Flash and V4-Flash-Vision-Exp. This article does not announce additional AvalAI redirects for those identifiers.
API request and response examples
These examples use an AvalAI API key stored in AVALAI_API_KEY. Responses below are illustrative, not live captures; identifiers, token counts, exchange rates, and costs vary. Base64 image data is truncated and cannot be decoded as shown.
DeepSeek V4.1 Flash
curl https://api.avalai.ir/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $AVALAI_API_KEY" \
-d '{
"model": "deepseek-v4.1-flash",
"messages": [{"role": "user", "content": "Give a concise checklist for testing a model migration."}]
}'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="deepseek-v4.1-flash",
messages=[
{
"role": "user",
"content": "Give a concise checklist for testing a model migration.",
}
],
)
print(response.choices[0].message.content)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: "deepseek-v4.1-flash",
messages: [{ role: "user", content: "Give a concise checklist for testing a model migration." }],
});
console.log(response.choices[0].message.content);Illustrative Chat Completions response; optional provider-specific fields may also be present:
{
"id": "chatcmpl-deepseek-v41-example",
"created": 1789084800,
"model": "deepseek-v4.1-flash",
"object": "chat.completion",
"choices": [
{
"index": 0,
"finish_reason": "stop",
"message": {
"role": "assistant",
"content": "Check prompt quality, tool round-trips, output formats, token limits, latency, and billing."
}
}
],
"usage": {
"prompt_tokens": 20,
"completion_tokens": 100,
"total_tokens": 120,
"prompt_tokens_details": {
"cached_tokens": 0
}
},
"estimated_cost": {
"unit": "0.0000630000",
"irt": 6.3,
"exchange_rate": 100000
}
}Grok 4.6
curl https://api.avalai.ir/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $AVALAI_API_KEY" \
-d '{
"model": "grok-4.6",
"messages": [{"role": "user", "content": "Plan an incident-triage dashboard and its verification steps."}]
}'The response uses the Chat Completions structure shown above, with the returned model identifier and usage for this request. Cost follows Grok 4.6's rates rather than the DeepSeek example.
GPT Image 2.5 generation
Use Flare for a general-purpose starting point, or change the model to gpt-image-2.5-sunburst for more precise creative work.
curl https://api.avalai.ir/v1/images/generations \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $AVALAI_API_KEY" \
-d '{
"model": "gpt-image-2.5-flare",
"prompt": "A studio product photo of a ceramic mug, soft natural lighting, plain cream background.",
"size": "1024x1024",
"quality": "medium",
"n": 1
}'Illustrative Images response; usage and additional fields depend on the route:
{
"created": 1789084800,
"data": [
{
"b64_json": "iVBORw0KGgoAAAANSUhEUg...[TRUNCATED]"
}
],
"usage": {
"input_tokens": 20,
"input_tokens_details": {
"text_tokens": 20,
"image_tokens": 0
},
"output_tokens": 439,
"total_tokens": 459
},
"estimated_cost": {
"unit": "0.0132700000",
"irt": 1327.0,
"exchange_rate": 100000
}
}This example combines $0.01317 of output-image tokens with $0.00010 of text-input tokens. It is not a guaranteed price for the sample prompt.
GPT Image 2.5 editing
Upload an existing image using the plural v1/images/edits endpoint. Let cURL set the multipart content type and boundary.
curl https://api.avalai.ir/v1/images/edits \
-H "Authorization: Bearer $AVALAI_API_KEY" \
-F "model=gpt-image-2.5-sunburst" \
-F "image=@product.png" \
-F "prompt=Change only the background to pale blue; preserve the mug, lighting, and composition." \
-F "size=1024x1024" \
-F "quality=high" \
-F "n=1"Editing returns the Images response structure. The request also incurs reference-image input costs; do not treat the output-only estimate as the total edit price.