One prompt returns a 1–15 second clip at 480p, 720p or 1080p, billed per second of finished video. Failed generations are never billed.
Cinematic concept shots and b-roll straight from a script.
Built for enterprise teams and API service operators. Unify chat, coding, image, and video workloads with centralized routing, keys, quotas, failover, usage, and cost governance.
Keep your existing SDKs and workflows — no business-code changes.
AnthropicChat / Coding
claude-fable-5-1 · claude-opus-5
claude-sonnet-5 · claude-haiku-4-5
10+ modelsPricing & details
OpenAIChat / Coding / Image
gpt-6-astra · gpt-5.6
codex · gpt-image-2
10+ modelsPricing & details
GoogleChat / Multimodal
gemini-3.8-flash · gemini-3.1-pro
gemini-3-pro-image
14+ modelsPricing & details
z.aiChat / Coding
glm-5.3
1M context
GLM seriesPricing & details
MoonshotChat / Reasoning
kimi-k3
1M context
Kimi seriesPricing & details
ByteDanceVideo / Image
seedance-2.5
seedream-5-0-pro
8+ modelsPricing & details
KuaishouVideo / Digital human
kling-v3-omni · kling-avatar
kling-lip-sync
6+ modelsPricing & details
DeepSeekChat
deepseek-v4-pro · flash
1M context
V4 seriesPricing & details
QwenChat
qwen3.8 max / flash · 3.7 max / plus / flash
1M context
3.8 / 3.7 seriesPricing & details
MiniMaxVideo
MiniMax-H3
768P / 2K · 4–15s
Hailuo seriesPricing & details
AlibabaVideo
wan3.0-video
happyhorse-1.1 · 480P–1080P
Wan · HappyHorsePricing & details
HopBase provides one stable control layer for routing, failover, access controls, and usage tracking across coding, API, image, and video workloads.
Handle batch workloads, shared development environments, and traffic spikes across business lines. Automatic multi-channel failover maintains availability, with concurrency sized to your workload.
Use stable, continuously maintained channels for Claude, Codex, OpenAI, and AWS Bedrock, built for long-running workloads and advanced reasoning.
Segment API keys, quotas, and billing by organization, project, team, or customer. Use the same controls for internal platforms and multi-tenant products.
Track every call by model usage for departmental chargebacks, customer billing, margin analysis, and usage audits.
The gateway is designed to leave your data nowhere to leak: content is never stored, credentials are scoped and revocable, and every call leaves an auditable trace.
Prompts and outputs are never persisted. Only billing metadata — model, token counts, latency, status — is kept for invoices and audits.
All traffic is TLS-encrypted in transit, and API keys are stored as one-way hashes — plaintext keys never touch the database.
Issue separate keys per company, project, or team with independent quotas and routing. Revoke any key instantly without touching the rest.
Model, usage, latency, and status are recorded per request — reconcilable bills, usage audits, and traceable anomalies, no black boxes.
Anthropic
Multi-line supply
Claude Max · AWS Bedrock direct · full-quota Max
OpenAI · Codex
Codex Plus / Pro
GPT-5.6 family · 1.05M context · image APIs
Gemini
Dual channels
Official direct · Azure channel · chat / image / video
GLM
GLM 5.3
New flagship · 1M context · thinking-length control
DeepSeek
deepseek-v4-flash
deepseek-v4-flash · 1M context
Qwen
qwen3.8-max
qwen3.8-max · 1M context
Seedance · Video / Image
Video & image
Seedance 2.5 / 2.0 video · Seedream image
Kling
Video · digital human
Video · image · digital human
MiniMax · Video
MiniMax-H3
New · text / image / reference to video · up to 2K
Kimi · Moonshot
kimi-k3 · 1M context
New · 1M context · cache reads from $0.30/M
Six production lines, picked by scenario — cinematic text-to-video, Bailian per-second clips, native-audio concept footage, e-commerce volume, digital-human presenters, and in-app sync clips.
One prompt returns a 1–15 second clip at 480p, 720p or 1080p, billed per second of finished video. Failed generations are never billed.
Cinematic concept shots and b-roll straight from a script.
Public reference images steer the subject and style of the clip; each reference is billed per image on top of the per-second rate.
Keep one product or character consistent across a whole series.
Aspect ratios from 1:1 and 16:9 to vertical 9:16 and 2:3. A five-second clip typically renders in one to two minutes.
Vertical cuts for shorts and horizontal masters from the same prompt.
Text, first / last frames, up to 10 reference images plus video and audio references drive one clip at 480P–1080P, billed per second of finished video.
Ad-grade footage with precise visual control.
Text-to-video, single-image first-frame, and 1–9 reference images share one per-second price across 480P–1080P.
Volume clip production at the lowest per-second rate.
Alibaba Model Studio contract on the original paths — official SDK code migrates by changing only the base URL and key. Failed generations are never billed.
Move existing DashScope workloads over in minutes.
One prompt produces a 4–15 second clip at 768P or 2K, with aspect ratios from 21:9 to 9:16. Failed generations are never billed.
Marketing b-roll and concept shots straight from a script.
Anchor the clip with a first and/or last frame image; motion is generated between them and the aspect follows your image.
Bring product stills and posters to life without a reshoot.
Reference images, video clips and audio guide the subject, motion and rhythm of the generated video.
Keep one character or style consistent across a whole series.
Async text-to-video and image-to-video tasks with multi-resolution output, billed only on completed clips.
High-volume short-video material produced through one stable API.
Overseas and domestic upstreams under one model ID, 480p–1080p standard clips with automatic failover.
Keep production running even when a single channel degrades.
Text-to-image, image-to-image and edits across Seedream 5.0 Pro / Lite and 4.5, up to 4K output.
Cover stills, posters and product shots from the same account.
1–5 portrait images plus one prepared audio track create a 1–20s video in 720p or 1080p.
E-commerce product briefings, batch-produced from one portrait and a script — no per-clip filming.
Detect faces in a source video, select one face, then sync it to a prepared audio track.
Relocalize finished videos for new languages and markets without reshooting.
Multi-language voice-over drives natural lip, expression, and head motion. Full-resolution output at 720p and 1080p, 1–20 seconds per clip, produced in batches.
One API call returns the finished 720p / 24 fps clip directly — no task queue, no polling.
Short clips generated inside interactive apps and agent workflows, at conversation speed.
Accepts text, images, and up to 10 seconds of video as input — edit or restyle an existing clip with one prompt.
The same model handles conversation and native video output with a 1M-token context window; the OpenAI-compatible chat entry stays available for text.
One OpenAI- and Anthropic-compatible endpoint works across coding agents, chat clients, and workflow platforms — change the base URL and keep everything else.
Straight answers on scale, reliability, pricing, and integration. See the docs for implementation details, or contact us about your workload.
Yes. HopBase provides one gateway for enterprises, one-person companies (OPCs), and teams across multiple business lines. Gateway-level groups, quotas, and cost tracking support batch jobs, shared team access, and production workloads. Concurrency is sized to your use case.
Most API proxies simply route requests through a single account, which limits reliability and operational control. HopBase adds high-concurrency routing, automatic multi-channel failover, scoped keys, quotas, and cost tracking for sustained production use.
HopBase keeps personal accounts out of your integration. You connect with enterprise API keys; AWS Bedrock requires no personal Claude login, and other Claude, Codex, and OpenAI channels are governed by group permissions.
HopBase supports Claude, OpenAI GPT and Codex, Gemini, GLM, Kimi, DeepSeek, Qwen, GPT Image, Gemini Image, Seedream, Seedance, Kling, and MiniMax families. AWS Bedrock currently supports Opus models only. See the public model catalog for the complete discovery list.
Claude Max is our most cost-effective line for frequent, cost-sensitive workloads. AWS Bedrock connects directly to AWS without a personal login and currently supports Opus models only. Claude Max (full quota) provides higher quotas for concurrent and long-running workloads. Our team recommends the right line based on budget, model coverage, and reliability requirements.
HopBase pricing is quoted one-on-one by our team based on the models you use, your monthly volume, and your scenario, with official list prices as the shared benchmark. Billing is always based on actual usage — contact us for a tailored quote.
No. Prepaid credits have no expiration date and remain available while your account is active.
HopBase provides one integration and one governance layer across providers. Keep Claude Code, Codex, and your OpenAI or Anthropic SDK while adding scoped keys, quotas, cost allocation, and usage auditing.
Official partner
One agent, commanded from WeChat and the tools your team already uses. Identical to the official offering via HopBase — a flat 10% off the official price.
Learn about WorkBuddyOfficial direct access to the next-gen omni-modal video model: one prompt to a 4–15s finished clip up to 2K, with native stereo audio and 11-language TTS.
Watch H3 samplesA new ecosystem partnership with Alibaba Cloud: cloud infrastructure and model resources working together for stable enterprise access — joint offerings rolling out soon.
Talk enterprise partnership
Official partner
From the team behind FastMoss, a top TikTok data provider: selection and creator data plus one-click viral-video cloning — Seedance from 50% off.
See the exclusive offer