# Classai Router

> One OpenAI-compatible API key for chat/coding models, text-to-speech, image generation and video generation, paid from one prepaid credit balance (Indonesia-based, prices in Rupiah). Failed requests are never charged.

Base URL: https://router.classai.id/v1
Auth: header `Authorization: Bearer sk_live_…` (create keys at https://router.classai.id/dashboard/api-keys). The same key works for every endpoint below.
MCP server (for Claude Code, Claude Desktop, Cursor, opencode…): https://router.classai.id/mcp — Streamable HTTP, same Bearer header. Tools: generate_image, edit_image, start_video, get_video, speak, list_voices, list_media_models.
OpenAPI spec: https://router.classai.id/openapi.json

## Endpoints

- POST /v1/chat/completions — OpenAI chat (stream: true supported). Also POST /v1/responses (OpenAI Responses) and POST /v1/messages (Anthropic format).
- GET /v1/models — chat models available to the key.
- POST /v1/audio/speech — text to speech {model, input, voice, response_format: mp3|wav, speed, style}. List engines and voices: GET /v1/audio/voices[?model=] (your cloned voices have cloned: true).
- POST /v1/audio/voices — clone a voice (multipart/form-data): file (sample: WAV/MP3/M4A/WebM, 3–60 s), consent (a recording of the voice's owner reading: "I, <name>, the owner of this voice, allow Classai to create an AI voice from my voice for the account <account email> on <date>."), agree=true, name, model (an engine with clone_fee_rp ≠ null, e.g. voxcpm2), optional owner, transcript. Whole request ≤ 10 MB. JSON alternative: audio + consent_audio as data URIs. → {object: "voice", id: "clone:<id>", name, model, created_at, sample_seconds, clone_fee_rp}; use the id as "voice" in /v1/audio/speech. Creating is free; engines that create the voice at the provider charge clone_fee_rp on its first use. Max 10 cloned voices per account, 6 new per hour. DELETE /v1/audio/voices/{id} removes one of yours.
- POST /v1/images/generations — images {model, prompt, aspect_ratio, resolution, quality (official OpenAI models: low|medium|high), n (1–4), image: [https URL | data URI] for references/edits (flux-kontext-* need exactly 1), response_format: b64_json|url}. Synchronous, 5 s – 4 min: use a client timeout ≥ 600 s.
- POST /v1/images/edits — OpenAI multipart edit (image / image[] files + prompt).
- POST /v1/videos — start a video {model, prompt, aspect_ratio, resolution, seconds, input_reference (first frame)} → {id, status: "in_progress"}. Options per model below; Veo is one price per video, the others are priced per second.
- GET /v1/videos/{id} — poll every ~10–15 s until status is "completed" (has url, valid 24 h) or "failed" (not charged). GET /v1/videos/{id}/content — the MP4.
- GET /v1/media/models — image & video models with prices (price_rp = from, price_estimate, price_per_second), options and waits.

## Errors

JSON {error: {message, type, code}}. 401 missing/invalid key · 402 not enough credits (top up at https://router.classai.id/kredit) · 403 key restricted to another model · 404 unknown model · 422 blocked by the safety filter (not charged) · 429 rate limit (retry after the Retry-After seconds; 60/min per key, 180/min per account, Enterprise 1,000 / 3,000; every chat answer has x-ratelimit-limit-requests / x-ratelimit-remaining-requests per key plus the -account pair, and GET /v1/account → limits shows the account's own) · 5xx temporary (not charged).

## Chat models (price in Rp per 1M input / output tokens)

- claude-opus-5.5 — Claude Opus 5.5 (Anthropic), 1000k context: Rp7.920 / Rp39.600
- claude-sonnet-5.5 — Claude Sonnet 5.5 (Anthropic), 1000k context: Rp3.960 / Rp19.800
- claude-opus-5 — Claude Opus 5 (Anthropic), 200k context: Rp9.900 / Rp49.500
- claude-sonnet-5 — Claude Sonnet 5 (Anthropic), 200k context: Rp3.960 / Rp19.800
- claude-sonnet-4.6 — Claude Sonnet 4.6 (Anthropic), 200k context: Rp5.945 / Rp29.700
- claude-haiku-4.5 — Claude Haiku 4.5 (Anthropic), 200k context: Rp990 / Rp4.950
- claude-haiku-5.5 — Claude Haiku 5.5 (Anthropic), 1000k context: Rp1.620 / Rp4.950
- glm-5.3 — GLM 5.3 (Zhipu GLM), 1000k context: Rp1.875 / Rp9.240
- deepseek-v4.1-flash — DeepSeek V4.1 Flash (DeepSeek), 1000k context: Rp1.250 / Rp4.540
- deepseek-v4.1-flash-1017 — DeepSeek V4.1 Flash 1017 (DeepSeek), 1000k context: Rp1.155 / Rp4.200
- kimi-k3 — Kimi K3 (Moonshot AI), 256k context: Rp2.970 / Rp14.850
- glm-5.2 — GLM 5.2 (Zhipu GLM), 1000k context: Rp1.850 / Rp9.240
- minimax-m3 — MiniMax M3 (MiniMax), 1000k context: Rp900 / Rp3.465
- kimi-k2.7-code — Kimi K2.7 Code (Moonshot AI), 250k context: Rp1.260 / Rp6.270
- minimax-m2.7 — MiniMax M2.7 (MiniMax), 200k context: Rp800 / Rp1.980
- deepseek-v4-flash — DeepSeek V4 Flash (DeepSeek), 1000k context: Rp1.150 / Rp4.640
- deepseek-v4-flash-0731 — DeepSeek V4 Flash 0731 (DeepSeek), 1000k context: Rp1.080 / Rp4.125
- deepseek-v4-pro — DeepSeek V4 Pro (DeepSeek), 1000k context: Rp3.135 / Rp8.250
- deepseek-v4-pro-0813 — DeepSeek V4 Pro 0813 (DeepSeek), 1000k context: Rp3.210 / Rp8.250
- glm-5.3-flash — GLM 5.3 Flash (Zhipu GLM), 1000k context: Rp1.110 / Rp4.620
- glm-5.1 — GLM 5.1 (Zhipu GLM), 200k context: Rp1.850 / Rp9.240
- kimi-k2.6 — Kimi K2.6 (Moonshot AI), 256k context: Rp1.255 / Rp6.270
- gemini-3.8-flash — Gemini 3.8 Flash (Google), 1000k context: Rp2.100 / Rp8.400
- claude-opus-4.6 — Claude Opus 4.6 (Anthropic), 200k context: Rp9.900 / Rp49.500
- gpt-6-astra — GPT-6 Astra (OpenAI), 400k context: Rp9.900 / Rp49.500
- gpt-5.6-sol — GPT-5.6 Sol (OpenAI), 400k context: Rp5.130 / Rp16.500
- gpt-6-sol — GPT-6 Sol (OpenAI), 1000k context: Rp5.130 / Rp16.500
- gpt-6.1-sol — GPT 6.1 Sol (OpenAI), 1100k context: Rp5.130 / Rp15.000
- gpt-5.6-terra — GPT-5.6 Terra (OpenAI), 1000k context: Rp4.320 / Rp12.375
- gemini-3.8-series — Gemini 3.8 Series (Google), 1000k context: Rp1.350 / Rp6.540
- gpt-5.6-luna — GPT-5.6 Luna (OpenAI), 1000k context: Rp1.860 / Rp9.285
- qwen3.8-max — Qwen 3.8 Max (Alibaba Qwen), 1000k context: Rp1.980 / Rp9.900
- claude-opus-4.8 — Claude Opus 4.8 (Anthropic), 1000k context: Rp9.900 / Rp49.500
- coding-china-auto — Coding Plan China (auto) (Classai), 1000k context: Rp1.875 / Rp9.240
- coding-us-auto — Coding Plan US (auto) (Classai), 200k context: Rp4.500 / Rp22.500
- gpt-6-luna — GPT-6 Luna (OpenAI), 1000k context: Rp1.860 / Rp9.285
- grok-4.7 — Grok 4.7 (xAI), 500k context: Rp3.170 / Rp9.900
- hy3 — Hy3 (Tencent Hunyuan), 256k context: Rp1.485 / Rp7.425
- qwen3.7-max — Qwen 3.7 Max (Alibaba Qwen), 1000k context: Rp1.465 / Rp7.305
- qwen3.7-plus — Qwen 3.7 Plus (Alibaba Qwen), 1000k context: Rp810 / Rp3.700
- claude-opus-4.7 — Claude Opus 4.7 (Anthropic), 200k context: Rp9.900 / Rp49.500
- kai-baik — Kai Grant · DeepSeek V4.1 Flash (DeepSeek), 1000k context: Rp0 / Rp0
- longcat-2.5:free — LongCat 2.5 (Free) (Meituan): Rp0 / Rp0
- step-3.7-flash:free — Step 3.7 Flash (Free) (StepFun): Rp0 / Rp0
- solar-mini4:free — Solar Mini 4 (Free) (Upstage): Rp0 / Rp0
- nemotron-3-ultra:free — Nemotron 3 Ultra 550B (Free) (NVIDIA), 1000k context: Rp0 / Rp0
- nemotron-3.5-super-vl:free — Nemotron 3.5 Super VL (Free) (NVIDIA), 1000k context: Rp0 / Rp0
- nemotron-3.5-lightning:free — Nemotron 3.5 Lightning (Free) (NVIDIA), 1000k context: Rp0 / Rp0
- nemotron-3-nano-omni:free — Nemotron 3 Nano Omni (Free) (NVIDIA), 256k context: Rp0 / Rp0
- north-mini-code:free — North Mini Code (Free) (Cohere), 256k context: Rp0 / Rp0
- gemma-4-31b:free — Gemma 4 31B (Free) (Google), 262k context: Rp0 / Rp0
- gemma-4-26b:free — Gemma 4 26B (Free) (Google), 262k context: Rp0 / Rp0
- dots-3-note:free — dots 3 Note (Free) (rednote hi lab), 512k context: Rp0 / Rp0
- flint-1.0:free — Flint 1.0 (Free) (Recursive), 262k context: Rp0 / Rp0
- lfm-3-1.7b:free — LFM 3 1.7B (Free) (Liquid AI), 262k context: Rp0 / Rp0

## Image models

- nano-banana-pro — Nano Banana Pro (Google): ≈ Rp3.000–Rp5.350 per image (billed by real token usage). Google's best image model: sharp text on images, faithful edits, up to 4K. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 1K/2K/4K; up to 8 reference images; ~15–40 s.
- nano-banana-2.1 — Nano Banana 2.1 (Google): ≈ Rp800–Rp1.700 per image (billed by real token usage). The newest Nano Banana: thinks before it draws, strong edits, up to 4K at a low price. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 1K/2K/4K; up to 8 reference images; ~10–25 s.
- nano-banana-2 — Nano Banana 2 (Google): ≈ Rp1.500–Rp3.350 per image (billed by real token usage). Fast, high quality and great at edits. Up to 4K. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 1K/2K/4K; up to 8 reference images; ~10–25 s.
- nano-banana-2-lite — Nano Banana 2 Lite (Google): ≈ Rp800–Rp800 per image (billed by real token usage). The lightest Nano Banana: quick drafts and simple edits for very few credits. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 1K; up to 4 reference images; ~5–10 s.
- nano-banana — Nano Banana (Google): ≈ Rp1.200–Rp1.200 per image (billed by real token usage). The original Nano Banana, still loved for natural-looking edits. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 1K; up to 4 reference images; ~8–15 s.
- gpt-image-2.5-flare — GPT Image 2.5 Flare (OpenAI): ≈ Rp500–Rp4.700 per image (billed by real token usage). OpenAI's newest image model, tuned for speed. Great prompt following and text. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 1K/2K; quality low/medium/high; up to 8 reference images; ~10–45 s.
- gpt-image-2.5-sunburst — GPT Image 2.5 Sunburst (OpenAI): ≈ Rp500–Rp4.700 per image (billed by real token usage). OpenAI's newest image model, tuned for precise edits that keep everything else untouched. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 1K/2K; quality low/medium/high; up to 8 reference images; ~15–150 s.
- gpt-image-2 — GPT Image 2 (OpenAI): ≈ Rp500–Rp18.650 per image (billed by real token usage). OpenAI's GPT Image 2: detailed posters and long prompts. High quality takes a few minutes. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 1K/2K; quality low/medium/high; up to 8 reference images; ~20–240 s.
- gpt-image-1.5 — GPT Image 1.5 (OpenAI): ≈ Rp500–Rp4.100 per image (billed by real token usage). OpenAI's GPT Image 1.5: reliable all-rounder for illustrations and edits. Aspect 1:1, 3:2, 2:3; resolution 1K; quality low/medium/high; up to 8 reference images; ~8–60 s.
- chatgpt-image — ChatGPT Image (OpenAI): ≈ Rp550–Rp4.250 per image (billed by real token usage). The image model used in ChatGPT, always the latest version. Aspect 1:1, 3:2, 2:3; resolution 1K; quality low/medium/high; up to 8 reference images; ~6–60 s.
- gpt-image-1 — GPT Image 1 (OpenAI): ≈ Rp500–Rp5.050 per image (billed by real token usage). OpenAI's first GPT Image model. Aspect 1:1, 3:2, 2:3; resolution 1K; quality low/medium/high; up to 8 reference images; ~10–60 s.
- gpt-image-1-mini — GPT Image 1 Mini (OpenAI): ≈ Rp500–Rp1.050 per image (billed by real token usage). The small, very cheap GPT Image model for drafts and volume. Aspect 1:1, 3:2, 2:3; resolution 1K; quality low/medium/high; up to 8 reference images; ~8–40 s.
- gpt-image-2.5 — GPT Image 2.5 Lite (OpenAI): Rp950 per image. Budget line of GPT Image 2.5: one flat price for any size up to 4K. Slower than the official models. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 1K/2K/4K; up to 4 reference images; ~30–150 s.
- mai-image-2.6 — MAI-Image 2.6 (Microsoft): Rp2.550 per image. Microsoft's image model: true-to-life photos with natural light and skin. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 1K; up to 5 reference images; ~20–40 s.
- mai-image-2.6-flash — MAI-Image 2.6 Flash (Microsoft): Rp1.300 per image. The faster, cheaper MAI-Image for everyday photos. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 1K; up to 5 reference images; ~12–20 s.
- seedream-5-pro — Seedream 5.0 Pro (ByteDance): Rp2.550 per image. Seedream's top model for complex instructions and the finest detail. Takes about 2 minutes. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 1K/2K; up to 10 reference images; ~110–140 s.
- seedream-5 — Seedream 5.0 (ByteDance): Rp1.100 per image. Natural Asian faces and product photos, high resolution. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 2K/3K; up to 10 reference images; ~30–40 s.
- seedream-5-flash — Seedream 5.0 Flash (ByteDance): Rp550 per image. Seedream's fast, low-cost version for drafts and volume. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 1K/2K; up to 10 reference images; ~15–20 s.
- seedream-4.5 — Seedream 4.5 (ByteDance): Rp1.250 per image. 4K images with crisp small text: posters, ads and UI mockups. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 2K/4K; up to 10 reference images; ~8–20 s.
- seedream-4 — Seedream 4.0 (ByteDance): Rp1.100 per image. The proven Seedream: consistent results and the cheapest 4K. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 1K/2K/4K; up to 10 reference images; ~7–15 s.
- flux-2-max — FLUX.2 Max (Black Forest Labs): Rp2.250 per image. FLUX's flagship: the most detailed photoreal images, up to 8 reference images. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 1K/2K; up to 8 reference images; ~8–15 s.
- flux-2-pro — FLUX.2 Pro (Black Forest Labs): Rp1.000 per image. Photorealistic images in seconds, consistent with reference images. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 1K/2K; up to 8 reference images; ~5–10 s.
- flux-2-flex — FLUX.2 Flex (Black Forest Labs): Rp1.900 per image. FLUX tuned for typography: logos, signs and text-heavy designs. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 1K/2K; up to 8 reference images; ~15–40 s.
- flux-2-klein — FLUX.2 Klein (Black Forest Labs): Rp500 per image. The small FLUX: images in a few seconds for the lowest price. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 1K/2K; up to 4 reference images; ~2–6 s.
- flux-1-dev — FLUX.1 Dev (Black Forest Labs): Rp650 per image. The classic open FLUX model, text to image only. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 1K; text only; ~5–8 s.
- flux-kontext-max — FLUX Kontext Max (Black Forest Labs): Rp2.150 per image. Edits one image by instruction with the best text editing. Needs an image to edit. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 1K; needs exactly 1 input image; ~8–15 s.
- flux-kontext-pro — FLUX Kontext Pro (Black Forest Labs): Rp1.100 per image. Edits one image by instruction: change the background, style or objects. Needs an image to edit. Aspect 1:1, 4:5, 3:4, 2:3, 9:16, 4:3, 3:2, 16:9, 21:9; resolution 1K; needs exactly 1 input image; ~6–12 s.

## Video models

- veo-3.1-fast — Veo 3.1 Fast (Google): Rp3.157/s at 720p, Rp3.788/s at 1080p (min Rp5.000 per video; e.g. Rp25.300 at defaults). Google's video model with sound. Fast and affordable, great for Reels and ads. With sound; 16:9, 9:16; 720p/1080p; 4/6/8 s; start image yes; ~1–2 min.
- veo-3.1 — Veo 3.1 (Google): Rp12.626/s at 720p, Rp12.626/s at 1080p (min Rp5.000 per video; e.g. Rp101.050 at defaults). Veo's highest quality: cinematic motion, realistic people and sound. With sound; 16:9, 9:16; 720p/1080p; 4/6/8 s; start image yes; ~1–3 min.
- seedance-2.5 — Seedance 2.5 (ByteDance): Rp3.245/s at 480p, Rp7.298/s at 720p, Rp18.427/s at 1080p (min Rp5.000 per video; e.g. Rp16.400 at defaults). ByteDance's newest video model: up to 30 seconds with sound, cinematic camera work. With sound; 16:9, 9:16, 1:1, 4:3, 3:4, 21:9; 480p/720p/1080p; 4–30 s (default 5); start image yes; ~2–6 min.
- seedance-2.0 — Seedance 2.0 (ByteDance): Rp2.219/s at 480p, Rp4.773/s at 720p, Rp11.812/s at 1080p (min Rp5.000 per video; e.g. Rp11.200 at defaults). Seedance 2.0: smooth motion and consistent characters, with sound, up to 1080p. With sound; 16:9, 9:16, 1:1, 4:3, 3:4, 21:9; 480p/720p/1080p; 4–15 s (default 5); start image yes; ~2–3 min.
- seedance-2.0-fast — Seedance 2.0 Fast (ByteDance): Rp1.410/s at 480p, Rp3.031/s at 720p (min Rp5.000 per video; e.g. Rp7.150 at defaults). The quicker Seedance 2.0 for drafts and social clips. With sound; 16:9, 9:16, 1:1, 4:3, 3:4, 21:9; 480p/720p; 4–15 s (default 5); start image yes; ~1–3 min.
- seedance-2.0-mini — Seedance 2.0 Mini (ByteDance): Rp1.112/s at 480p, Rp2.387/s at 720p (min Rp5.000 per video; e.g. Rp5.650 at defaults). The cheapest Seedance with sound: great value for many short clips. With sound; 16:9, 9:16, 1:1, 4:3, 3:4, 21:9; 480p/720p; 4–15 s (default 5); start image yes; ~1–2 min.
- wan-3.0 — Wan 3.0 (Alibaba): Rp1.105/s at 480p, Rp2.210/s at 720p, Rp4.419/s at 1080p (min Rp5.000 per video; e.g. Rp5.550 at defaults). Alibaba's Wan 3.0: native soundtrack, up to 30 seconds, the lowest price per second. With sound; 16:9, 9:16, 1:1, 4:3, 3:4; 480p/720p/1080p; 2–30 s (default 5); start image yes; ~2–3 min.
- wan-3.0-prime — Wan 3.0 Prime (Alibaba): Rp1.896/s at 480p, Rp3.792/s at 720p, Rp7.583/s at 1080p (min Rp5.000 per video; e.g. Rp9.500 at defaults). Wan 3.0 on priority hardware: same model, faster results. With sound; 16:9, 9:16, 1:1, 4:3, 3:4; 480p/720p/1080p; 2–30 s (default 5); start image yes; ~1–2 min.
- wan-2.7 — Wan 2.7 (Alibaba): Rp2.528/s at 720p, Rp3.157/s at 1080p (min Rp5.000 per video; e.g. Rp12.650 at defaults). Wan 2.7 text- and image-to-video at 720p or 1080p, with sound. With sound; 16:9, 9:16, 1:1, 4:3, 3:4; 720p/1080p; 2–15 s (default 5); start image yes; ~1–3 min.
- happyhorse-1.1 — HappyHorse 1.1 (Alibaba): Rp3.119/s at 720p, Rp4.718/s at 1080p (min Rp5.000 per video; e.g. Rp15.600 at defaults). Alibaba's HappyHorse: lifelike motion that keeps the subject steady. 16:9, 9:16, 1:1, 4:3, 3:4; 720p/1080p; 3–15 s (default 5); start image yes; ~1–3 min.
- minimax-h3 — MiniMax H3 (MiniMax): Rp903/s at 768p (min Rp5.000 per video; e.g. Rp5.450 at defaults). MiniMax's H3 (Hailuo family): stereo sound always on, up to 15 s. Slow but cheap. With sound; 16:9, 9:16, 1:1, 4:3, 3:4, 21:9; 768p; 4–15 s (default 6); start image yes; ~3–10 min.
- oxygen-1.0 — Oxygen 1.0 (AZ8): Rp602/s at 480p, Rp602/s at 768p (min Rp5.000 per video; e.g. Rp5.000 at defaults). Budget video model with a native soundtrack, same price at 480p and 768p. With sound; 16:9, 9:16; 480p/768p; 4–15 s (default 6); start image yes; ~1–3 min.

## Voice engines (price per ~1,000 characters ≈ 1 minute)

- voxcpm2 — VoxCPM2 (OpenBMB): Rp800. Natural Indonesian voices, cloned from a short clip. Runs on Classai GPUs.
- dots-tts — dots.tts (rednote): Rp800. Very close voice clones from a ~3-second clip. Runs on Classai GPUs.
- elevenlabs-v4 — Eleven v4 (ElevenLabs): Rp4.200. ElevenLabs' most expressive model. Steer it with emotion tags like [whispers] or [laughs].
- elevenlabs-v4-turbo — Eleven v4 Turbo (ElevenLabs): Rp2.100. A faster Eleven v4 at half the price, for high volume and bots.
- elevenlabs-flash — Eleven Flash v2.5 (ElevenLabs): Rp2.100. ElevenLabs' fastest model (~75 ms), great for bots and live replies.
- elevenlabs-v2 — Eleven Multilingual v2 (ElevenLabs): Rp4.200. The steady, proven ElevenLabs model for long narration and audiobooks.
- openai-tts — OpenAI TTS (OpenAI): Rp1.414. OpenAI's 9 multilingual voices. Quick, consistent and affordable for bots and drafts.
- openai-tts-hd — OpenAI TTS HD (OpenAI): Rp2.829. The higher-quality OpenAI voices: cleaner audio for narration, videos and podcasts.

## Embeddings, rerank & moderation (price in Rp per 1M input tokens)

- POST /v1/embeddings — OpenAI format {model, input: string | string[] (≤ 2,048), dimensions?, encoding_format: float|base64} → {data: [{embedding, index}], usage}.
- POST /v1/rerank — {model, query, documents: string[] (≤ 1,000), top_n?} → {results: [{index, relevance_score, document}], usage} sorted best first. Billed on total tokens (the query counts once per document).
- POST /v1/moderations — OpenAI format {model: omni-moderation-latest, input: string | string[]} → {results: [{flagged, categories, category_scores}]}.
- Cost = tokens × price, rounded up to the smallest unit; failed requests are not charged. Docs: https://router.classai.id/docs/embeddings

- text-embedding-3-small — Embedding 3 Small (OpenAI), 1536 dims: Rp942,86 per 1M tokens. Cheap, good default for search and RAG. Supports dimensions.
- text-embedding-3-large — Embedding 3 Large (OpenAI), 3072 dims: Rp6.128,57 per 1M tokens. Most accurate OpenAI embedding. Supports dimensions.
- text-embedding-ada-002 — Embedding Ada 002 (OpenAI), 1536 dims: Rp4.714,29 per 1M tokens. Legacy model, for existing indexes built with it.
- bge-m3 — BGE-M3 (BAAI), 1024 dims: Rp471,43 per 1M tokens. Multilingual (Indonesian included), 8K context, lowest cost. No dimensions option.
- bge-reranker-v2-m3 — BGE Reranker v2 M3 (BAAI): Rp471,43 per 1M tokens. Multilingual reranker: scores documents against a query (0–1).
- omni-moderation-latest — Omni Moderation (OpenAI): Rp9.428,57 per 1M tokens. Flags harassment, hate, self-harm, sexual, violence and illicit content.

## Docs

- [Quick start](https://router.classai.id/docs/mulai)
- [Endpoints](https://router.classai.id/docs/endpoint)
- [Voice API](https://router.classai.id/docs/suara)
- [Image API](https://router.classai.id/docs/gambar)
- [Video API](https://router.classai.id/docs/video)
- [Credits](https://router.classai.id/docs/paket)
- [Error codes](https://router.classai.id/docs/error)
- [Tool integrations](https://router.classai.id/docs/integrasi)

## Product pages

- [Voice AI](https://router.classai.id/voice) · [Image AI](https://router.classai.id/image) · [Video AI](https://router.classai.id/video) · [Models & prices](https://router.classai.id/model) · [Credits](https://router.classai.id/kredit)