Get started

Models

Every model ID you can call, and how to choose between them.

Pass one of the model IDs below in the model field of your request. New models land as they ship, and live pricing for each is in the dashboard.

Choosing a model

  • Claude models (claude-*) are best used with Claude Code. For direct requests, call them through Chat Completions; the Messages API currently serves Claude Code.
  • GPT models (gpt-*) work with any OpenAI-compatible client: the Chat Completions and Responses APIs, the OpenAI SDK, Codex CLI, chat apps and coding agents.
  • Open-weight models work through Chat Completions, for direct requests and agent frameworks. deepseek-v4.1-flash-1017 also runs in Claude Code.

For an agent framework, use an open-weight or GPT model. See the model support matrix.

Claude models

Call these from Claude Code, or through /v1/chat/completions for direct requests.

Model IDContextBest for
claude-fable-5-21MThe newest Fable, the premium flagship.
claude-fable-5-11MThe previous Fable, at the top of the lineup.
claude-fable-51MA pinned earlier Fable snapshot.
claude-opus-5-51MThe newest Opus, for the hardest reasoning and agentic work.
claude-opus-51MThe previous Opus, for demanding reasoning and agentic work.
claude-opus-4-81MTop-tier reasoning and coding.
claude-opus-4-71MA pinned earlier Opus snapshot.
claude-opus-4-61MA pinned earlier Opus snapshot.
claude-sonnet-5-51MThe newest Sonnet, a strong, fast default.
claude-sonnet-51MThe previous Sonnet, fast and capable.
claude-sonnet-4-61MA pinned Sonnet snapshot, fast and balanced.
claude-haiku-5-51MThe newest Haiku, the fastest and most economical option.
claude-haiku-4-5200KThe previous Haiku, fast and light.

All text models support tool use. See Tool use and Reasoning.

GPT models

Call these through the OpenAI-compatible /v1/chat/completions or /v1/responses endpoints.

Model IDContextBest for
gpt-6-astra1MThe flagship, for the most demanding agentic work.
gpt-6.1-sol1MThe newest Sol, fast and economical for everyday agentic work.
gpt-6.1-terra1MThe newest Terra, a balanced mid-tier option for reasoning, computer use and writing.
gpt-6-luna1MThe lightest and most economical option.
gpt-5.6-sol1MThe previous Sol, a pinned agentic model.
gpt-5.6-terra1MThe previous Terra, a pinned mid-tier model.

GPT models support tool calling and reasoning. See GPT (OpenAI-compatible) to connect the OpenAI SDK, Cursor, or curl.

Open-weight models

Call these through the OpenAI-compatible /v1/chat/completions endpoint. deepseek-v4.1-flash-1017 also runs in Claude Code, over /v1/messages; the others aren't reachable through /v1/messages, and none through /v1/responses. They are the recommended choice for opencode.

Model IDContextBest for
deepseek-v4-flash-07311MThe cheapest model on this endpoint: a fast open-weight model from DeepSeek.
deepseek-v4.1-flash-10171MDeepSeek's newer V4.1 Flash, improved for coding and agent work. Also runs in Claude Code.
qwen3.8-max1MAlibaba's flagship Qwen model.
kimi-k31MMoonshot AI's Kimi model.
glm-5.3256KZ.ai's agent-oriented GLM, built for reasoning and tool use.
glm-5.3-flash1MA low-cost GLM tier. Natively multimodal, with the larger context of the two GLM tiers.

Image models

Call these through /v1/images/generations and /v1/images/edits.

Model IDBest forSizesReference images
gpt-image-2.5OpenAI's GPT Image 2.5: high quality generation and editing.1K, 2KUp to 10
nano-banana-2.1Google's Nano Banana 2.1: generation and editing up to 4K.1K, 2K, 4KUp to 14
seedream-proByteDance's Seedream 5.0 Pro: photorealistic generation and editing.1K, 2KUp to 10
seedream-flashByteDance's Seedream 5.0 Flash: fast, economical generation and editing, lowest cost per image.1K, 2KUp to 10

Every image model supports text-to-image and all ten aspect ratios. Sizes are per model: nano-banana-2.1 renders at 1K, 2K or 4K, the others at 1K or 2K.

Every image model also edits. /v1/images/edits takes reference images on all four, up to the limit in the table above; sending more returns 400. Use them for edits, multi-image blends and subject consistency, and cite each in the prompt as @image1, @image2, and so on. See the Seedream guide.

Video models

Call these through the async /v1/videos endpoint.

Model IDBest forInputDurations
seedance-2.5Long clips with synchronized audio, and multi-image references.Text, image, or references5, 8, 12, 16, 20, 24, 30s
seedance-2.0The lower-cost Seedance: clips up to 15s with synchronized audio, and multi-image references.Text, image, or references5, 8, 10, 12, 15s
wan-3.0Alibaba's Wan 3.0: long clips up to 1080p with synchronized audio.Text, image, or references5, 8, 12, 16, 20, 24, 30s
gemini-omni-1.1-flashGoogle's Gemini Omni 1.1 Flash: short 720p clips with synchronized audio.Text, image, or references4, 6, 8, 10s

Durations are per model: a value one model accepts, another rejects. Every clip from every video model includes synchronized audio.

Every video model takes references. How many input_reference files you send picks the mode:

  • None: text-to-video from the prompt alone.
  • One: animates that image as the first frame.
  • Two or more: uses them as subject and style references rather than frames, up to 30 on seedance-2.5, 9 on seedance-2.0, 20 on wan-3.0 and 10 on gemini-omni-1.1-flash. Cite them in the prompt as @image1, @image2, and so on.

All four also accept an optional last frame alongside exactly one input_reference, for first plus last frame animation. See Videos.

seedance-2.5 renders at 480p, 720p or 1080p in six aspect ratios (16:9, 9:16, 4:3, 3:4, 1:1, 21:9; every size is under Sizes), and generates synchronized audio at no extra cost. Clips run up to 30 seconds, and price scales with both duration and resolution. A single starting frame is cover-cropped to the size you order. It also makes quick 480p drafts you can finish in 1080p: see Drafts.

seedance-2.0 takes the same sizes as seedance-2.5 at a lower price, with clips up to 15 seconds and synchronized audio at no extra cost. A single starting frame is cover-cropped to the size you order.

wan-3.0 renders at 480p, 720p or 1080p in 16:9, 9:16, 4:3, 3:4 or 1:1, every ratio of a resolution at one price, with clips up to 30 seconds. With a single starting frame there is no aspect setting: the clip takes its shape from the frame, and the size only picks the resolution tier.

gemini-omni-1.1-flash renders at 720p, landscape (1280x720) or portrait (720x1280), with clips up to 10 seconds. Like seedance-2.5, it cover-crops a single starting frame to the size you order.

A prompt is required in every mode. When you supply a starting frame it should describe the motion, not the scene: the scene is already the frame you uploaded.

Deprecated models

These models are still served until the date below, then removed. After that, a request to one is rejected and the error names its replacement. Sizes and durations differ between a model and its replacement, so check the tables above when you switch.

Model IDReplacementServed until
nano-banana-2nano-banana-2.12026-11-13
nano-banana-progpt-image-2.52026-11-13
flux-uncensoredseedream-flash2026-11-13
pixverse-v6wan-3.02026-11-13
wan-spicywan-3.02026-11-13

GET /v1/models marks each one with "deprecated": true, plus its "replacement" and "retires_on" date, so you can check for them in code. Every response for one of these models also carries the standard Deprecation and Sunset headers, a Link to this section, and x-roteo-model-replacement with the model to move to.

Pricing

Live per-token and per-image prices for every model are on the pricing page.

Listing models in code

Call GET /v1/models (no key needed) for the OpenAI-style JSON list (?type=video adds each video model's sizes, ratios, reference limits and the price of every size and duration), or point an agent at /v1/llms.txt for the live models and endpoints in plain text.

Models | Roteo