Skip to main content
As of November 6, 2026, this page lists the last 30 days of LaoZhang API launches, price changes, retirements, and route status, newest first. The tag on each entry tells you what kind of change it is. Action needed
  • Claude Sonnet 4.5 was withdrawn on November 4: replace claude-sonnet-4-5-20250929 with claude-sonnet-5-5. See replacements
  • Calling DeepSeek, GLM, or Kimi over the Claude protocol? Since November 4 this needs a Claude Code group token; default tokens get a 503. See how to switch
  • gpt-image-1.5 and gpt-image-1-mini stop on December 1: move to GPT Image 2.5 or 2.6. Compare migration targets
Recently launched
  • November 5: GPT Image 2.6 is live and returns animated GIFs and SVG vectors from one request. See prices and requests
  • October 30: Meta Muse video generation is live, with a soundtrack in every clip, from $0.30 for 10 seconds and now up to 60 seconds. See pricing and usage

​
Latest announcements

​
November 6, 2026

  • New feature Call GPT Image by its official model names per call: new GPT Image VIP group, $0.03 per image with native 4K If your code or tool only accepts official model names, switch the token’s group to GPT Image VIP and leave model unchanged. The official GPT Image 2, 2.5 and 2.6 model names then run on the VIP route, exactly like the matching -vip models. See groups and billing
    • In the default group, the 2.5 and 2.6 official model names are now also billed per call at $0.03 on the VIP route; gpt-image-2 in default is unchanged
    • The official-forward groups are unchanged, and the same model names are still billed by tokens there
    • SVG and the 2.6 xhigh / max tiers aren’t available; use official forwarding for those
  • Minor update GPT Image 2.6 VIP now returns animated GIFs at $0.03 per call, from generation and edits. See sizes and an example

​
November 5, 2026

  • New model GPT Image 2.6 is live: images in as little as 8 seconds, animated GIFs and SVG vectors from one request With an official-forward group token, set model to gpt-image-2.6-flare or gpt-image-2.6-sunburst. Billing is per token at OpenAI’s price, the same rates as 2.5. SVG writing is free with the default gpt-6.1-terra and eight other low-cost text models; other writing models are billed at their own rate. See per-image costs and the GIF and SVG requests
    • high quality and above costs the same as 2.5, about $0.053 per 1024 image, and Flare returns it in 8–12 s
    • "output_format": "gif" returns a looping animation; 512×512 with nine frames costs about $0.06
    • default group tokens can use gpt-image-2.6-flare-vip and gpt-image-2.6-sunburst-vip at $0.03 per image, up to high quality, without SVG (GIF output added November 6)
    • default group tokens can use the web route gpt-image-2.6-web at $0.03 per call; it returns GIFs with frame control but not SVG

​
November 4, 2026

  • Removed Claude Sonnet 4.5 is withdrawn: switch to claude-sonnet-5-5 or claude-sonnet-4-6 claude-sonnet-4-5-20250929 was withdrawn on November 4, and requests for it now return 503. Anthropic has deprecated the model and retires it from its API on November 30. Change only the model; the endpoint and other parameters stay the same. See replacements and prices
    • claude-sonnet-5-5: Anthropic’s recommended replacement, input $2 and output $10, cheaper than Sonnet 4.5
    • claude-sonnet-4-6: the same generation, at the same price as Sonnet 4.5
  • New feature Claude Code token group now open: one key for Claude, DeepSeek, GLM and Kimi in Claude Code Choose the Claude Code group when you create a token, and Claude Code, CC Switch, the Anthropic SDK and other Claude-protocol tools can call Claude, DeepSeek, GLM and Kimi at the same prices as default. See which tools use it and how to switch
    • OpenAI-compatible requests (Chat Completions, Responses) keep using the default group
    • Tools that use both formats (OpenClaw, Hermes Agent, CC Switch) need two tokens
    • default tokens can no longer call DeepSeek, GLM or Kimi over the Claude protocol; multi-turn tool calls for these models over the OpenAI-compatible format work again
  • New feature GPT-6 now calls tools in Chat Completions: use it in Cline and Cherry Studio Chat Completions requests that sent tools to GPT-6 used to fail with Function tools with reasoning_effort are not supported. The same requests now return tool calls, at the same price and with no code changes. See the models and an example
    • Covers GPT-6, GPT-6.1, and GPT-5.6 reasoning models; send reasoning_effort together with tools
    • These models also accept max_tokens in Chat Completions
  • New feature Muse style and camera presets: anime, fantasy, short-drama, and product-ad looks in one field Stop describing the look and camera in every prompt: add one field, at the same price ($0.30 for 10 seconds, $1.55 for 60). Without it the output is the same as before, so existing code keeps working. See recipes by use case and a sample clip
    • 7 looks, including anime, Chinese fantasy CG, Western comic, cyberpunk, product footage, and live-action film
    • 13 camera techniques, including short-drama hooks, product-ad close-ups, fantasy speed ramps, whip pans, and subtle acting

​
November 2, 2026

  • Restored GPT Image -vip is stable again, with native 4K The three per-call GPT Image VIP models are stable again: from 16:00 on November 1 to 23:00 on November 2 (UTC+8), every hour succeeded above 96%, 99.4% overall. If you moved to the web route or official forwarding in late October, change model back to -vip. See how to switch back and spot upscaled 4K
    • 4K is rendered natively at the requested size, not upscaled or super-resolved, and every size from 1K to 4K costs $0.03 per image
    • Many per-call services return “4K” that has the right size but upscaled detail; check the image at 100% and run the downscale test to tell them apart
  • New model Muse video adds 40, 50, and 60-second models muse-video-40s, muse-video-50s, and muse-video-60s are live. Each generates one continuous 40–60 second clip or extends an existing video. Call them the same way as muse-video-20s; each video has a fixed price. See prices and continuation rules
    • muse-video-40s: 40 seconds, $1.05 per video
    • muse-video-50s: 50 seconds, $1.30 per video
    • muse-video-60s: 60 seconds, $1.55 per video
    When continuing a video, the finished clip is as long as the model name says, source included, and the source must be at least 2 seconds shorter. For example, muse-video-20s takes sources up to 18 seconds.

​
October 30, 2026

  • Removed grok-imagine-image is withdrawn The earlier Grok Imagine model grok-imagine-image was withdrawn on October 30, and calls to it now return 503. Change model to grok-imagine-image-2.0; the endpoints and aspect ratios stay the same. See how to migrate
  • New model Meta Muse video generation is live Standard token groups can call it through /v1/videos for text-to-video, first and last frames, reference images, and video continuation, with a soundtrack in every clip. Each video has a fixed price, and the token needs the Usage first or Per-call billing mode. See pricing and how to call it
    • muse-video: 5, 8, or 10 seconds, $0.30 each
    • muse-video-20s: 20 seconds, $0.55 each
    • muse-video-30s: 30 seconds, $0.80 each

​
October 28, 2026

  • Removed whisper-1 is withdrawn The whisper-1 transcription model was withdrawn on October 28, and requests for it now return 503. Change model to gpt-4o-mini-transcribe; the endpoint is unchanged. Requests that asked for subtitles or timestamps need a parameter change too. See how to migrate
    • whisper-1 → gpt-4o-mini-transcribe for everyday transcription, gpt-4o-transcribe for higher accuracy

​
October 25, 2026

  • New model Kimi K3.1 Pro, MiniMax M3, DeepSeek Flash and 4 more models are live Seven text models are available in the default group through the OpenAI-compatible /v1/chat/completions, billed per token, with reasoning returned in reasoning_content. See prices and usage
    • Kimi: kimi-k3.1-pro, kimi-k3.1-pro-highspeed
    • MiniMax: MiniMax-M3, MiniMax-M3.1-Flash
    • Zhipu: glm-5.2-fast-preview, glm-5.3-prime
    • DeepSeek: deepseek-flash (same model and price as deepseek-v4-flash)
  • Removed deepseek-v3.2 is retired DeepSeek no longer serves deepseek-v3.2, so LaoZhang API has retired the model ID and requests to it now fail. Change model to deepseek-flash. Read more

​
October 24, 2026

  • New model MiniMax H3 video generation is live MiniMax-H3 is available in the default, vip, and svip groups for text-to-video, first and last frames, and reference images, video, and audio, with a soundtrack in every clip. It costs $0.05 per second at 768P and $0.08 at 2K, and the token needs the Usage first or Usage-based billing mode. See pricing and how to call it

​
October 23, 2026

​
October 18, 2026

  • Restored Claude models are back Eleven Claude models, including Opus 5.5, Sonnet 5.5, Haiku 5.5, and Fable 5.1, are listed again at Anthropic’s list prices. Call them from the default and claude_code token groups in the Anthropic Messages or OpenAI-compatible format. Output is capped at 32,768 tokens per response, and PDF document input and Anthropic server tools are not available. See models, prices, and limits

​
October 14, 2026

  • Price change New Seedance prices Seedance 2.0 and 2.5 now cost 1.2× Volcengine’s list price, and most rates drop about 5%: Seedance 2.0 at 720p goes from $8.28 to $7.88 per 1M tokens, for example. Seedance 2.5 at 1080p now uses Volcengine’s 1080p rate, $13.20 per 1M tokens. Requests and tokens don’t change. Compare old and new prices
  • New model Removed Wan 3.0 launch, new Wan video prices, and Wan 2.6 retirement wan3.0-video and the faster wan3.0-video-prime are live in the Wan group, and one model covers text, image, and reference to video plus video editing. Wan video now costs 1.2× Alibaba Cloud’s list price: Wan 2.7 is $0.1028 per second at 720P and $0.1714 at 1080P. The four wan2.6-* models are retired, so switch to Wan 2.7 or Wan 3.0. See the new prices and how to migrate
  • New model Gemini 4 Argon launch gemini-4-argon, which Google announced on September 30, is live with no Google account or early-access request needed. It costs $2 input and $10 output per 1M tokens, the same as Google’s launch price, and thinking tokens are billed as output. It works through both the OpenAI-compatible and Gemini-native endpoints. See prices, thinking levels, and examples

​
October 13, 2026

  • New model GPT-6.1 Sol launch gpt-6.1-sol works through Chat Completions and Responses at $2 input and $10 output per 1M tokens, the same as OpenAI. Cache reads cost $0.1, half the gpt-6-sol rate. Tool calling requires the Responses API. See prices, differences from GPT-6 Sol, and examples

​
October 7, 2026

  • New model Nano Banana 2.1 launch Google’s update to Nano Banana 2, gemini-nano-banana-2.1, is live at $0.045 per call for 1K, 2K, and 4K alike. That is $0.01 less than Nano Banana 2, with better text and layout. Existing Nano Banana 2 code switches by changing the model ID; 512px output isn’t supported. See what changed and how to switch
Announcements from before October 7 are in the 2026 announcement archive.
Current models and prices are listed in the model and pricing catalog. Your token groups and actual charges are shown in the LaoZhang API console and call logs.

​
Frequently asked questions

​
Where should I confirm whether a newly announced model is available?

Read the announcement, then check the model and pricing catalog and the console model and pricing page for your token group. Send a small request with the production token before increasing traffic.

​
Are prices and service states in archived announcements still current?

Not necessarily. Archived entries describe the state when they were published, and later entries record the changes that superseded them. Use the latest announcements, the model and pricing catalog, and your call logs for current availability, pricing, and billing.

​
Archives

​
Get updates