Available models

• gpt-6-astra: GPT-6 flagship (phased rollout by OpenAI; availability is scoped per API key, check GET /v1/models)
• gpt-5.6-sol / gpt-5.6-terra / gpt-5.6-luna: flagship / balanced / lightweight
• gpt-5.5: complex reasoning and Codex
• gpt-5.4: balanced performance
• gpt-5.4-mini: low latency and cost
• gpt-5.3-codex-spark: lightweight Codex tasks
• codex-auto-review: automated code review

• gpt-image-2: Image 2.0
• gpt-image-2.5-flare: Image 2.5, everyday default
• gpt-image-2.5-sunburst: Image 2.5, higher fidelity and slower at the same settings

Use POST /v1/images/generations and /v1/images/edits. GPT Image 2 supports auto, 1K, 2K, and 4K sizes; GPT Image 2.5 adds custom WIDTHxHEIGHT sizes and five quality tiers from low to max. See Image API.

• gemini-omni-flash-preview: text, image, and video input; 3-10 second video output
• gemini-3.6-flash: newest Flash model; currently under observation
• gemini-3.1-pro-preview
• gemini-3.1-pro-preview-customtools
• Other Gemini 2.5-3.5 Pro, Flash, and Flash-Lite models

Standard chat uses POST /v1/chat/completions; Omni's complete multimodal API uses POST /v1beta/interactions.

• seedream-5-0-pro: 1K / 2K text-to-image, image-to-image, and editing
• seedream-5-0-lite: lightweight 2K / 3K / 4K generation and editing
• seedream-4-5: 2K / 4K generation and editing with JPEG output

Uses the OpenAI Images protocol. Signed image URLs are valid for 24 hours.

• gemini-2.5-flash-image: Banana, 1K
• gemini-3-pro-image*: Banana Pro family, 1K / 2K / 4K
• gemini-3.1-flash-image*: Banana 2 family, 1K / 2K
• gemini-3.1-flash-lite-image: Banana 2 Lite, 1K

* abbreviates a family here and is not a requestable model ID. Use the complete ID returned by GET /v1/models.

• glm-5.3
• glm-5.2
• glm-5.2-fp8

Uses OpenAI Chat Completions. In client advanced settings, choose "OpenAI Chat Completions (routing required)."

• kimi-k3: Moonshot AI flagship model; 1M context, up to 32,768 output tokens

Serves POST /v1/chat/completions only. Claude Code and Codex CLI work through CC Switch local routing, which translates their protocols. See Kimi K3.

• claude-fable-5-1: Anthropic's most capable model; 1M context, cache hits at $0.25/M
• claude-fable-5: previous-generation Fable
• claude-opus-5: next-generation Opus
• claude-opus-4-8
• claude-opus-4-7
• claude-opus-4-6
• claude-sonnet-5: next-generation Sonnet
• claude-sonnet-4-6
• claude-sonnet-4-5-20250929, short name claude-sonnet-4-5
• claude-haiku-4-5-20251001, short name claude-haiku-4-5

• dreamina-seedance-2-5-260628: Seedance 2.5 (Overseas); 480p / 720p; integer durations from 4-30 seconds or -1; seven supported ratios
• doubao-seedance-2-0-260128-a: China standard, 480p / 720p / 1080p
• doubao-seedance-2-0-fast-260128-a / doubao-seedance-2-0-mini-260615-a: China fast / mini, 480p / 720p
• doubao-seedance-2-5-260628-a: Seedance 2.5 (China), 480p / 720p / 1080p, same durations and ratios as overseas 2.5
• dreamina-seedance-2-0-*: overseas standard; existing clients may retain the three complete standard IDs when switching to a China key
• dreamina-seedance-2-0-fast-*: overseas fast
• dreamina-seedance-2-0-mini-*: overseas mini

China and overseas variants use separate group keys. Use the complete ID returned by /v1/models. See Video generation for integration and pricing.

• kling-v3-omni, kling-v3, kling-v3-turbo, kling-o1
• kling-v2-6, kling-v2-5-turbo, kling-v2-1, kling-v2-0, kling-v1-6
• kling-v3-motion-control, kling-v2-6-motion-control, kling-avatar, kling-lip-sync
• kling-image-v3, kling-image-v3-omni, kling-image-o1, kling-image-v2-1, kling-image-v2-1-i2i, kling-image-v2-1-multi-ref, kling-image-expand

These IDs use the Tencent VOD Kling contract and asynchronous task polling. Configure the provider with SecretId, SecretKey, and SubAppId; see Kling video and image API for the exact images[] / videos[] schema, reference limits, scenes, and billing buckets.

• MiniMax-H3: 4-15 s, 768P / 2K, up to 9 reference images, 3 video clips, and 3 audio tracks
• MiniMax-H3-Max: fast variant, 5-15 s, 480P / 768P, text and first / last frame only

Every clip ships with native stereo audio. Uses the official MiniMax v2 contract over POST /v1/video/generate; see MiniMax Hailuo H3.

• wan3.0-video: 2-30 s omni-reference clips, first / last frame, reference image, video, audio, document, or web link
• happyhorse-1.1-t2v / -i2v / -r2v: 3-15 s text, first-frame, or reference-image generation

These keep Alibaba Model Studio's own DashScope paths, so official Alibaba code migrates by changing only the base URL and the key. See Wan 3.0 and HappyHorse.

• grok-4.6, grok-4.5, grok-4.3: xAI chat and reasoning
• grok-4.20-0309-reasoning, grok-4.20-multi-agent-0309
• grok-imagine-image, grok-imagine-image-2.0, grok-imagine-image-quality: image generation
• grok-imagine-video-1.5: text-to-video with optional reference images, billed per second

See Grok for the chat and image contracts, long-context pricing, reasoning-token billing, and server-side tool charges. Video runs on the asynchronous task API in its own plan group; see Video generation.

• deepseek-v4-pro-202606: Tencent TokenHub self-deployed official release (V4 Pro 0813)
• deepseek-v4-flash-202605: Tencent TokenHub self-deployed official release (V4 Flash 0731)

Both models share one plan group and are called over OpenAI Chat Completions. Only these two dated IDs are production models: deepseek-v4-flash is rejected with a 400 naming its replacement, and deepseek-v4-pro returns 404 model_not_found. See DeepSeek V4.

DeepSeek bills peak and off-peak official rates, where off-peak is half of peak. HopBase quotes and bills a single all-day rate derived from the peak list price and does not follow the off-peak reduction.