Changelog

• Added eleven_v3 and eleven_v4 text-to-speech API references, audio task polling, and the audio model guide.
• Added Audio pricing, including provider-reported character-cost billing.

• Added Kling4.0 and gemini-omni-1.2-flash to the video model guide, API reference, and pricing; refreshed the chat model catalog and pricing.
• Corrected token-based starting-price units to USD per 1M tokens and refreshed billable-item pricing.
• Replaced unavailable gpt-5.3-codex examples and marked the historical Sora 2 reference as unavailable.

• Added gpt-image-2.6-flare, gpt-image-2.6-sunburst, and grok-imagine-image-2.0 to the image API Reference and model catalog.
• Added gpt-6.1-terra, gemini-4-argon, kimi-k3.1-pro, and glm-5.3-flash to the Chat model catalog; gpt-6.1-terra is also listed for Responses.
• Refreshed request parameters and public pricing, including Seedream 5.0 Pro layer decomposition and transparent backgrounds, Gemini image aspect ratios, and video input limits.
• Removed sora-2 from current model recommendations because it is no longer returned by the public model catalog. Its historical API Reference remains available; the first-call example now uses seedance-2-0.

• Updated Chat Completions, Responses, Anthropic Messages, and Gemini GenerateContent error schemas to use their protocol-specific error envelopes and HTTP error statuses.
• Added error handling guidance for balance and API key quota failures, upstream failures, and errors after SSE streaming starts.
• Clients using top-level code/msg for these compatibility endpoints should migrate to protocol error fields. Task-based generation APIs retain their endpoint-specific contracts.

• Added the OpenAI Decisions API reference for gpt-6-luna, including predicate, choice, and score questions, per-question refusals, and token usage.
• Updated the model catalog with Decisions endpoint support and clarified that its input and cache rates follow standard Luna pricing while output tokens are not charged under OpenAI’s Decisions pricing rules.

​

New models

• gemini-omni-flash-preview adds Google’s preview multimodal video generation model through POST /videos/generations.
• Added grok-4.5 and grok-4.6 to the Chat model catalog.
• Updated Video pricing and Chat pricing for these models.

​

Image parameter updates

• gemini-nano-banana-2.1 now lists aspect ratios in WxH form, such as 21x9 and 9x16, matching the current upstream schema.
• gpt-image-2.5-flare and gpt-image-2.5-sunburst now list the upstream size presets and auto option.

​

New image model

• gemini-nano-banana-2.1 adds image generation and conversational editing with 1K–4K output, multi-image references, and search grounding.
• Updated Image pricing with the new model’s token rates.

​

Chat catalog refresh

• Added gpt-6.1-sol, gemini-3.8-flash, claude-haiku-5-5, and claude-sonnet-5-5 to Models and Chat pricing.
• Removed gpt-5.3-codex from the catalog and pricing page because it is no longer returned by the public sources; refreshed the Responses API model example.
• Removed the zero-price placeholder from the MiniMax-H3 API description.

​

New video model

• wan3.0-video adds Alibaba audiovisual generation with text-to-video, first/last-frame image-to-video, up to 1080P and 30-second output, and optional synchronized audio.
• Video pricing is billed by generated milliseconds and varies by resolution; see Video pricing for the current rates.

​

Chat catalog refresh

• Added gpt-6-luna, gpt-6-sol, claude-fable-5, claude-fable-5-1, claude-opus-5, claude-opus-5-5, and claude-sonnet-5.
• Removed gemini-2.5-pro, which is no longer returned by the current public model catalog.
• Refreshed Chat pricing for the updated catalog.

​

Schema updates

• model is now explicitly required for gemini-2.5-flash-image, seedance-2-0, and wan2.6.
• gpt-image-2.5-flare and gpt-image-2.5-sunburst no longer document the unsupported response_format field.

​

GPT Image 2 schema update

• gpt-image-2 now documents extends.edit.mask for local image editing; transparent mask areas are regenerated and the field is used together with images.
• The unified image-generation endpoint continues to require model as its routing key.

​

Gemini video understanding

• Added a video understanding with Gemini guide covering upload session creation, direct file upload, processing status, native generateContent, and cleanup.
• Added API Reference pages for the Gemini-only file lifecycle and native multimodal request.
• Gemini file URIs are not reusable across OpenAI, Anthropic, generation, or compatibility endpoints.
• Validated the workflow end to end with gemini-3.1-flash-lite and added portable filenames, complete upload headers, gateway error details, and automatic cleanup guidance.

​

New image models

• gpt-image-2.5-sunburst — precision-focused image generation and editing with up to 16 reference images, mask-guided edits, custom dimensions, and transparent backgrounds.
• gpt-image-2.5-flare — fast image generation and editing with the same reference-image, mask, custom-size, quality, and transparent-output controls.
• Both models support text-to-image and image-to-image workflows through POST /v1/images/generations; cached text input starts at $0.9375 / 1M tokens.

​

Catalog and pricing updates

• Added gpt-6-astra to the Chat model catalog and pricing page.
• gemini-3.1-flash-lite-image — starting price drops from 0.25 → 0.1875 USD / token.
• Removed sora-2 from the video pricing table because it is no longer returned by the current public pricing source; its API Reference page remains available.

​

New video models

• seedance-2-0-fast — adds a low-latency Seedance option for 4–15 second multimodal video generation, with image, video, and audio references.
• seedance-2-0-mini — adds a lightweight Seedance option for high-volume video generation and editing workflows.

​

Updated schemas and model catalog

• seedance-2-5 — intelligent duration (-1) is now the default; reference inputs are exposed consistently as images, videos, and audios.
• MiniMax-Hailuo-02 — prompt is now optional, matching image-conditioned generation workflows.
• Added glm-5.2 and glm-5.3 to the current Chat model catalog.
• Removed the stale gemini-3-pro-image-preview and gemini-3.1-flash-image-preview pages because these models are no longer returned by the public catalog.

​

Resource-oriented generation API

• Use POST /v1/images/generations and GET /v1/images/generations/{task_id} for image generation.
• Use POST /v1/videos/generations and GET /v1/videos/generations/{task_id} for video generation.
• Use public GET /v1/models to discover the currently visible model catalog without a Bearer API key.
• Existing /v1/image-generation, /v1/video-generation, and historical query routes remain supported. See Legacy API compatibility for the complete mapping.
• Output asset URLs are dynamically signed. Avoid long-term URL caching, and do not blindly retry create requests after an uncertain timeout.

​

Model catalog sync

• Added gemini-3-pro-image, gemini-3.1-flash-image, and MiniMax-H3 from the current upstream catalog.
• Removed the stale seedance-v1.5-pro reference because it is no longer returned by the visible model source.
• Refreshed starting prices for seedream-4.5 (0.029040 USD/image), seedream-5.0-lite (0.030305 USD/image), seedream-5.0-pro (0.045000 USD/call), and wan2.6 (0.304850 USD/call).

​

Seedance 2.0 schema update

• seedance-2-0 — adds an images array with up to 9 reference inputs.
• prompt now accepts up to 2,500 characters, reduced from 5,000. Update clients that validate or send longer prompts.

​

New model

• seedance-2-5 — ByteDance’s Seedance 2.5 supports text-to-video and image-to-video generation up to 30 seconds, synchronized audio, and image, video, or audio references through the unified video generation API. Starting from 6.08 USD / token.

​

Seedance 2.0 parameter update

• seedance-2-0 — prompt now accepts up to 5,000 characters.
• duration now supports 14s and 15s generations.
• resolution now includes 1080P and 4K options.
• prompt is no longer marked as required, matching image-to-video and media-conditioned workflows.

​

Schema upgrade — richer model parameter docs

• Enum dropdowns for size, resolution, quality — Playground renders select boxes instead of free-text inputs.
• Default values shown next to each field (e.g. seedance-2-0.resolution defaults to 720P).
• Conditional rules spelled out where they apply — e.g. on minimax-hailuo-02 the resolution field now documents the three rules that constrain it (duration, first_frame_image, last_frame_image).
• Field ordering reflects how parameters are typically filled in, not alphabetical.
• gpt-image-2 — previously missing the required model field in the schema; now present.

​

Chat endpoints — removed unsupported params

• OpenAI Chat Completions: model, messages, stream, max_tokens, stop
• Anthropic Messages: model, messages, max_tokens, system, stream, stop_sequences

​

Docs site refresh

​

Pricing updates

• kling-v3 — starting price drops from 0.0740 → 0.0610 USD / second (~17.6% lower).
• gemini-3.1-flash-image-preview — starting price adjusted to 0.0503 USD / call (was 0.0500). This historical preview model is no longer available.
• gpt-image-2 — starting price now listed: 0.9375 USD / token.
• seedance-2-0 — starting price now listed: 3.87 USD / token.

​

New model

• happyhorse-1.0 — Alibaba’s HappyHorse 1.0, supporting both text-to-video and image-to-video through POST /v1/video-generation. Starting from 0.1158 USD / second.

​

Updated models

• gpt-image-2 — now accepts up to 10 reference images for image-to-image.
• seedance-2-0 — adds new audios, first_frame_image, last_frame_image, and videos inputs.

​

Heads up — API base URL changed

​

New models

• gpt-image-2 — OpenAI’s next-gen image model, now available through the unified POST /v1/image-generation endpoint.
• seedance-2-0 — Bytedance Seedance 2.0, available through POST /v1/video-generation.

​

Other updates

• Pricing — kling-v3 now bills per second of generated video instead of per call.
• Changed — sora-2 now accepts a resolution parameter. Minor field-description refreshes on the historical gemini-3-pro-image-preview model, kling-v3, and minimax-hailuo-02.

​

Introducing LinkModel

​

What’s available today

• 6 video models — Sora 2, Kling v3, MiniMax Hailuo 2.3, MiniMax Hailuo 02, Wan 2.6, Seedance v1.5 Pro
• 6 image models — Gemini 3.1 / 3 Pro / 2.5 Flash, Seedream 4.0 / 4.5 / 5.0 Lite
• Two unified endpoints — POST /v1/video-generation and POST /v1/image-generation
• One response envelope — { code, data, msg, request_id } across all endpoints

​

Get started

1. Create an API key in your Dashboard
2. Browse all models, or jump into the API Reference
3. Copy any endpoint’s cURL example, drop in your key, and run

​

Pricing