Latest AI image models

ChatGPT Images 2.5

OpenAI’s announcement on September 8, 2026 says Images 2.5 is in ChatGPT, ChatGPT Work, and Codex on desktop, mobile, and web, on every tier. OpenAI says it keeps reference subjects more reliably, follows multi-turn edits more reliably, and cuts generation latency by up to 50% compared with Images 2.0. In chat, the same post adds Sketch (draw a reference), templates for formats such as flyers and product photos, comments on an image, and prompt sharing.

The API splits that release in two. GPT Image 2.5 Flare (gpt-image-2.5-flare, snapshot gpt-image-2.5-flare-2026-09-08) is the faster everyday model. GPT Image 2.5 Sunburst (gpt-image-2.5-sunburst, snapshot gpt-image-2.5-sunburst-2026-09-08) is the one OpenAI points to when edit precision matters more than speed. Both take text and image input and output an image. OpenAI bills the API per token. The Sunburst model card lists image output at $30 per million tokens on standard processing and $15 per million tokens on batch. That is a token rate, not a price per picture.

Sources, read October 25, 2026: Introducing ChatGPT Images 2.5 and the Sunburst and Flare model cards.

Midjourney V8.2

Midjourney’s version docs say the default is V8.2, released as the default on July 24, 2026. Midjourney describes it as an aesthetics, image quality, and Personalization update, and says the new Edit Model replaces Omni Reference, Character Reference, and the Retexture tool.

V8.1 came out on April 14, 2026 and was the default from June 10 to July 23, 2026. Midjourney says standard V8.1 jobs render about 4 to 5 times faster than earlier versions, and that HD makes native 2K images without a separate upscale. On that docs page, HD costs 1.3 GPU minutes and standard costs 0.8 GPU minutes. Those minute figures are written under V8.1, not repeated as a V8.2 price.

Niji 7, the anime line made with Spellbrush, launched on January 9, 2026. Midjourney says it is more coherent and more literal than earlier Niji versions.

Source, read October 25, 2026: Midjourney version docs.

Google Nano Banana

Google’s image models in Gemini now use the Nano Banana name. The posts we could read, in date order:

• Nano Banana 2 (Gemini 3.1 Flash Image), February 26, 2026. Google says it keeps up to five characters and up to 14 objects consistent, renders from 512px to 4K, and can render and translate text inside the image. It rolled out in the Gemini app, Search, AI Studio, the Gemini API, Flow, and Google Ads.
• Nano Banana 2 Lite (gemini-3.1-flash-lite-image), June 30, 2026. Google calls it the fastest, cheapest Gemini image model in that post, says a text-to-image result takes about 4 seconds, and lists $0.034 per 1K image. That price is Google’s API figure for a 1K image, not a Gemini app subscription price. The same post still lists Nano Banana Pro (Gemini 3 Pro Image) for work where accuracy matters more than speed, and the original Nano Banana (Gemini 2.5 Flash Image) as the legacy model.
• Nano Banana 2.1, named on October 20, 2026. Google’s small-business Gemini post calls 2.1 “our latest image generation model” and pairs it with uploaded websites, brand kits, fonts, and color palettes. That post does not give a parameter count, an API model id, or a per-image price, so those fields stay blank here.

Sources, read October 25, 2026: Nano Banana 2, Nano Banana 2 Lite, and Gemini tools for small business owners.

FLUX 3 Image Klein

Black Forest Labs’ site now leads with FLUX 3, a line that covers image, video, and audio. The image release we could date is FLUX 3 Image Klein, October 15, 2026. BFL says it is a 5.6 billion parameter open-weights model for text-to-image, editing, and up to 10 reference images, with native 4K output and bounding boxes that place an element or limit an edit. Weights come as a base model and a distilled model. The FLUX Kommunity License, as BFL describes it, allows free commercial use for individuals and for companies under $5 million in annual revenue. Larger companies need a self-hosted license or the API.

BFL’s own hardware notes: generator weights are 11.2 GB in BF16 and 5.6 GB in FP8, and the INT8 build runs on NVIDIA GeForce RTX cards with 8 GB of VRAM. BFL says the distilled INT8 model makes a 1 megapixel image in about 4 seconds on an RTX 5090 and in under 9 seconds on an RTX 4070, through ComfyUI. API pricing on that post starts at $0.009 per 1K image. BFL also says Klein led its own internal tests. That ranking is theirs, not a BestAIImageGen score.

Source, read October 25, 2026: FLUX 3 Image Klein.

Ideogram 4.0

Ideogram released 4.0 on June 3, 2026 as an open-weight text-to-image model. The technical post says it is 9.3 billion parameters. The press release says it renders multilingual text, can place a logo, headline, or subject with a bounding box, and outputs photoreal images at 2K. Hosted access is on ideogram.ai and the API. Weights are downloadable. Ideogram’s model picker docs still label 4.0 as the latest Ideogram model. License tiers are on Ideogram’s licensing page. We are not copying a price from a third-party table.

Sources, read October 25, 2026: Ideogram 4.0 press release and available models.