Models on ATP Token
Text, image, video, audio and embeddings, all behind one key. Open any model for its rates, capabilities and quickstart.
Rate table69 models
Editor's picks
Hand-picked by our editors — not usage rankings.
Trending
New
- Claude Fable 5Anthropic's capability ceiling for long-running agents — classifier refusals can fall back.
- GPT-5.6 TerraBalanced GPT-5.6 tier for support, internal tools, and document analysis at volume.
- Qwen 3 8 MAXAvailable on the ATP gateway — see the model page for rates and caps.
- Qwen 3 7 MAXAvailable on the ATP gateway — see the model page for rates and caps.
Text34
Text models for chat, coding agents, and tool-using workflows. Each page lists USD rates and how access is enforced per project.
Maximum-performance GPT-5.4 tier — route only the hardest calls here.
Anthropic's capability ceiling for long-running agents — classifier refusals can fall back.
GPT-5.6 flagship — frontier reasoning, long-horizon agents, ultra subagent mode.
Finishes the same agent and coding work with fewer tokens than GPT-5.4.
Opus 4.x flagship — agentic coding with parallel-subagent workflows and effort control.
Opus tier for hard engineering, high-res visual analysis, and polished slides and docs.
Balanced GPT-5.6 tier for support, internal tools, and document analysis at volume.
Sonnet all-rounder — computer-use automation and 1M-context coding sessions.
OpenAI frontier workhorse for coding agents, computer use, and document-heavy work.
Available on the ATP gateway — see the model page for rates and caps.
Google's Pro-class preview — deep reasoning and precise tool use over a 1M-token context.
Most agentic Sonnet — end-to-end coding with browser and terminal autonomy.
Google's multimodal workhorse — four thinking levels, tuned for agentic coding at Flash cost.
Alibaba's closed-weights Max flagship — agentic coding, CLI automation, Chinese–English work.
Available on the ATP gateway — see the model page for rates and caps.
Available on the ATP gateway — see the model page for rates and caps.
Available on the ATP gateway — see the model page for rates and caps.
Smallest, fastest GPT-5.6 — bulk summarization, drafting, routine automation.
Fastest Claude tier — high-volume chat, extraction, and subagent execution.
GLM's long-horizon flagship — autonomous engineering runs of up to 8 hours per Z.ai.
DeepSeek's flagship MoE — default thinking mode and 1M context at open-model rates.
Newest GLM flagship — 1M-token context for project-scale coding agents, open weights.
Moonshot's open-weight 1T MoE — swarm-style agents and long autonomous runs.
Kimi's coding specialist — long-horizon software engineering with leaner thinking tokens.
Z.ai's frontier MoE base tier — agentic engineering at the lowest GLM list price.
Google's budget Flash preview — Pro-grade reasoning at under a quarter of Pro cost.
Available on the ATP gateway — see the model page for rates and caps.
Available on the ATP gateway — see the model page for rates and caps.
Available on the ATP gateway — see the model page for rates and caps.
Available on the ATP gateway — see the model page for rates and caps.
Available on the ATP gateway — see the model page for rates and caps.
DeepSeek's throughput tier — bulk workloads at ATP's lowest text list rate.
Available on the ATP gateway — see the model page for rates and caps.
Available on the ATP gateway — see the model page for rates and caps.
Image9
Image generation and multimodal image meters on the gateway catalog.
Gemini 3 Pro Image · quality tier for in-image text, grounding, 2K/4K.
Nano Banana Pro editing · localized edits with lighting and camera control.
Gemini 3.1 Flash Image · final-quality renders up to 4K at Flash speed.
Nano Banana 2 editing · subject-consistent edits and in-image text localization.
Gemini 2.5 Flash Image · fast text-to-image drafts, one price across tiers.
Nano Banana editing · natural-language edits on your source images.
OpenAI image model with a thinking step — dense text and product shots.
Available on the ATP gateway — see the model page for rates and caps.
Available on the ATP gateway — see the model page for rates and caps.
Video24
Video models (often preview) with resolution or token meters — contact sales for volume.
Kling O3 video-to-video · 1080P restyles of existing footage.
Kling O3 video edit · targeted natural-language edits at 1080P.
HappyHorse 1.1 text-to-video · native synced audio and multilingual lip-sync.
HappyHorse 1.1 image-to-video · stills to clips that ship with native audio.
HappyHorse 1.1 reference-to-video · up to 9 reference images for consistent casts.
HappyHorse 1.0 video edit · rework a 3–60 s clip with up to 5 references.
Kling O3 video-to-video · restyle existing footage at 720P.
Kling O3 video edit · natural-language object and background changes, 720P.
Kling 3.0 text-to-video · 1080P pro tier for commercial-grade masters.
Kling 3.0 image-to-video · 1080P pro tier for image-anchored masters.
Kling O3 unified multimodal · 1080P text-to-video quality tier.
Kling O3 image-to-video · 1080P delivery-grade key-art animation.
Kling O3 reference-to-video · consistent subjects at 1080P.
Kling O3 reference-to-video · up to 7 references at 1080P.
Wan 2.7 text-to-video · plans composition before rendering multi-shot clips.
Wan 2.7 image-to-video · first-frame conditioned motion at 720/1080P.
Kling 3.0 text-to-video · 720P standard tier for drafts and social clips.
Kling 3.0 image-to-video · animate stills at the 720P standard tier.
Kling O3 unified multimodal · 720P text-to-video for fast iteration.
Kling O3 image-to-video · 720P keyframe-conditioned motion drafts.
Kling O3 reference-to-video · keep subjects consistent at 720P.
Kling O3 reference-to-video · up to 7 reference images at 720P.
Seedance 2.0 quality tier — cinematic multi-shot clips up to 1080p.
Seedance fast tier — quicker 480/720p renders for prompt iteration.
Audio1
Audio / speech models billed per character or listed meter.
Embedding1
Embedding models for retrieval and semantic search under the same project keys.