Back to changelog

v2026.5.1: New image models + Gemini catalog sync

Three new image-generation families are live — OpenAI gpt-image-2, Google Imagen 4 (Fast / Standard / Ultra), and Nano Banana Pro. Gemini catalog updated against Google's live model list. Bug fix: gateway no longer leaks OpenAI-only fields to Gemini.

osmAPI v2026.5.1 — OpenAI gpt-image-2 + Google Imagen 4

At a glance

Category What's new
OpenAI image generation gpt-image-2, gpt-image-2-2026-04-21
Google Imagen 4 imagen-4.0-fast-generate-001, imagen-4.0-generate-001, imagen-4.0-ultra-generate-001
Google Gemini image nano-banana-pro-preview
Google Gemini text / embeds gemini-3.1-flash-lite (GA), gemini-embedding-2 (GA)
Legacy Gemini (deprecated) gemini-2.0-flash, gemini-2.0-flash-lite — Google shutdown 2026-06-01
Bug fix modalities field no longer forwarded to Gemini

All new models work through the standard POST /v1/chat/completions endpoint. No SDK or client changes required.

OpenAI — gpt-image-2

OpenAI's flagship image model, accepting text and image inputs and returning image output. Available immediately on osmAPI under both its rolling alias and dated snapshot.

Model Released
gpt-image-2 2026-04-21
gpt-image-2-2026-04-21 2026-04-21

Pricing — per 1M tokens (both models identical):

Token type Standard Cached
Text input $5.00 $1.25
Image input $8.00 —
Image output $30.00 —

Tip: include an image_url part in the user message to edit an existing image instead of generating from scratch.

Google Imagen 4

Three quality tiers of Google's Imagen 4, billed per generated image (no token pricing).

Model Price per image Best for
imagen-4.0-fast-generate-001 $0.02 High-throughput, lowest cost
imagen-4.0-generate-001 $0.04 Standard quality, balanced cost
imagen-4.0-ultra-generate-001 $0.06 Highest fidelity, premium quality

Imagen uses Google's :predict endpoint with a different request/response shape than Gemini chat. The gateway handles the translation, so callers continue to use POST /v1/chat/completions with model: "imagen-4.0-..." and receive a standard OpenAI-style response.

Google Gemini — new entries

Image

Model Pricing Notes
nano-banana-pro-preview Text in $2/1M · Text out $12/1M · Image out $120/1M Google's marketing alias for gemini-3-pro-image-preview — same model

Text & embeddings

Model Input (per 1M) Output (per 1M) Notes
gemini-3.1-flash-lite $0.25 $1.50 GA promotion of the preview
gemini-embedding-2 $0.20 — GA promotion of the preview

Deprecated by Google (action required by 2026-06-01)

Model Input (per 1M) Output (per 1M) Shutdown date
gemini-2.0-flash $0.10 $0.40 2026-06-01
gemini-2.0-flash-lite $0.075 $0.30 2026-06-01

Google will stop serving these on 2026-06-01. Both models still respond today, but the gateway now flags them as deprecated and will auto-filter them from routing after the cutoff. Migrate to gemini-2.5-flash or gemini-3.1-flash-lite for similar pricing and behavior.

Bug fix — Gemini image 400 errors

Before this release, a small percentage of requests to Gemini image models (e.g. gemini-3-pro-image-preview) failed with:

HTTP 400 — Invalid JSON payload received. Unknown name "modalities": Cannot find field.

The root cause was the gateway forwarding the OpenAI-style modalities request field to every provider, including Google AI Studio. Gemini's REST schema doesn't accept that field.

The fix scopes the forwarding to OpenAI only. gpt-audio models still receive modalities and audio parameters as before; Gemini requests are now clean.

Compatibility

No caller changes required. All new code paths are scoped to imageGenerations: true + the matching provider, so existing flows are untouched:

  • Gemini chat-image (gemini-2.5-flash-image, gemini-3-pro-image-preview, gemini-3.1-flash-image-preview) — unchanged
  • Alibaba qwen-image-* — unchanged
  • Z.AI cogview-* — unchanged

All six new image models were verified live against the upstream providers before release.