Models & credit rates

TokenSmith connects one OpenAI-compatible API to several major providers plus open-weight hosts, so you can pick models without juggling separate accounts. Text rows list input and output credits per 1M tokens and typical max context. Image rows list credits per generated image; video rows list credits per video job (async create + poll). Usage debits your credit balance; see Billing for packs and tiers.

Text models use POST /v1/chat/completions; image models use POST /v1/images/generations; video models use POST /v1/videos and GET /v1/videos/{id} to poll jobs. Fal Kling motion-control models use POST /v1/{fal-model-path}/motioncontrol. Video billing is per second of requested length (seconds or duration in the body). Together SKUs below were originally quoted for 5s clips; the table shows the equivalent credits per second. Usage debits your balance; see Billing for top-ups.

Closed source models

Text models

Pick a model in POST /v1/chat/completions; we route it to the right backend.

Model In credits/1M Out credits/1M Max ctx
gpt-5.4 5,000 30,000 270K
gpt-5.4-mini 1,500 9,000 270K
gpt-5.4-nano 400 2,500 270K
Model In credits/1M Out credits/1M Max ctx
claude-opus-4-7 10,000 50,000 200K
claude-sonnet-4-6 6,000 30,000 200K
claude-haiku-4-5 2,000 10,000 200K
Model In credits/1M Out credits/1M Max ctx
grok-4.20-0309-reasoning 4,000 12,000 2M
grok-4.20-0309-non-reasoning 4,000 12,000 2M
grok-4-1-fast-reasoning 400 1,000 2M
Model In credits/1M Out credits/1M Max ctx
gemini-3.1-pro 4,000 24,000 200K
gemini-3-flash 1,000 6,000 1M

Google may use different tiers for very long prompts. See Gemini API pricing.

Image generation models

Video models

OpenAI POST /v1/videos; debit scales with seconds (default 4s if omitted). Google Veo uses the same create path; the gateway forwards to Gemini predictLongRunning and bills from parameters.durationSeconds (default 8) and parameters.resolution. Fal Kling regular video also uses POST /v1/videos; motion control uses a multipart route shown below.

Model ID Credits / sec Resolution Notes
sora-2 200 720p Sora 2 (with audio)
Model ID Credits / sec Resolution Notes
veo-3.1-generate-preview 800 720p Standard (with audio)
800 1080p Standard (with audio)
1,200 4K Standard (with audio)
veo-3.1-fast-generate-preview 200 720p Fast (with audio)
240 1080p Fast (with audio)
600 4K Fast (with audio)
veo-3.1-lite-generate-preview 100 720p Lite (with audio); 4K output not supported
160 1080p Lite (with audio)
Model ID Credits / sec Endpoint Notes
fal-ai/kling-video/v2.6/pro/text-to-video 280 POST /v1/videos Kling 2.6 Pro T2V · native audio default; 140 with generate_audio=false
fal-ai/kling-video/v2.6/pro/image-to-video 280 POST /v1/videos Kling 2.6 Pro I2V · native audio default; 140 without audio, 336 with voice control
fal-ai/kling-video/v2.5-turbo/standard/image-to-video 84 POST /v1/videos Kling 2.5 Turbo Standard I2V
fal-ai/kling-video/o1/image-to-video 224 POST /v1/videos Kling O1 Pro first-frame / last-frame I2V
fal-ai/kling-video/o1/standard/image-to-video 168 POST /v1/videos Kling O1 Standard first-frame / last-frame I2V
fal-ai/kling-video/o1/standard/reference-to-video 168 POST /v1/videos Kling O1 Standard reference I2V
fal-ai/kling-video/o1/video-to-video/reference 336 POST /v1/videos Kling O1 Pro reference V2V
fal-ai/kling-video/o1/video-to-video/edit 336 POST /v1/videos Kling O1 Pro video edit
fal-ai/kling-video/o1/standard/video-to-video/edit 252 POST /v1/videos Kling O1 Standard video edit
Model ID Credits / sec Endpoint Notes
fal-ai/kling-video/v2.6/standard/motion-control 140 POST /v1/fal-ai/kling-video/v2.6/standard/motioncontrol Multipart: motion video + character image
fal-ai/kling-video/v2.6/pro/motion-control 224 POST /v1/fal-ai/kling-video/v2.6/pro/motioncontrol Higher quality motion transfer
fal-ai/kling-video/v3/standard/motion-control 252 POST /v1/fal-ai/kling-video/v3/standard/motioncontrol Kling 3.0 Standard motion control

Voice models

Open models (cloud) Open weights

Open-weight text SKUs use POST /v1/chat/completions; image SKUs use POST /v1/images/generations. Use each string as the model field.

Text models

Model ID In credits/1M Out credits/1M Max ctx
openai/gpt-oss-120b 300 1,200 128K
OpenAI/gpt-oss-20B 100 400 128K
Model ID In credits/1M Out credits/1M Max ctx
google/gemma-4-31B-it 400 1,000 256K
google/gemma-3n-E4B-it 120 240 32K
Model ID In credits/1M Out credits/1M Max ctx
Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8 4,000 4,000 256K
Qwen/Qwen3-Coder-Next-FP8 1,000 2,400 262K
Qwen/Qwen2.5-7B-Instruct-Turbo 600 600 32K
Model ID In credits/1M Out credits/1M Max ctx
zai-org/GLM-5.1 2,800 8,800 200K
zai-org/GLM-5 2,000 6,400 202K
Model ID In credits/1M Out credits/1M Max ctx
moonshotai/Kimi-K2.5 1,000 5,600 262K
Model ID In credits/1M Out credits/1M Max ctx
MiniMaxAI/MiniMax-M2.7 600 2,400 229K
MiniMaxAI/MiniMax-M2.5 600 2,400 192K
Model ID In credits/1M Out credits/1M Max ctx
deepseek-ai/DeepSeek-V3.1 1,200 3,400 128K
deepseek-ai/DeepSeek-R1 6,000 14,000 160K
Model ID In credits/1M Out credits/1M Max ctx
meta-llama/Llama-3.3-70B-Instruct-Turbo 1,760 1,760 128K
Model ID In credits/1M Out credits/1M Max ctx
deepcogito/cogito-v2-1-671b 2,500 2,500 32K
Model ID In credits/1M Out credits/1M Max ctx
LiquidAI/LFM2-24B-A2B 60 240 32K
Model ID In credits/1M Out credits/1M Max ctx
essentialai/rnj-1-instruct 300 300 32K

Image generation models

Call POST /v1/images/generations with model, prompt, and supported options. GPT Image 2 also supports multimodal POST /v1/responses requests with image inputs. Per-image credits are listed below; token-billed GPT Image 2 jobs may vary by input and output size.

Model ID Credits / image Notes
ByteDance-Seed/Seedream-3.0 36 Seedream 3.0
black-forest-labs/FLUX.2-pro 60 FLUX.2 Pro
openai/gpt-image-1.5 68 GPT Image 1.5
gpt-image-2 100+ GPT Image 2 · Responses + image generation
google/flash-image-3.1 89 Nano Banana 2 (Gemini Flash image)
Qwen/Qwen-Image-2.0-Pro 150 Qwen Image 2.0 Pro
google/gemini-3-pro-image 250 Nano Banana Pro (Gemini image)

Video models

Same POST /v1/videos flow; debit scales with seconds (default 5s if omitted, matching the legacy 5s list prices). Values below are credits per second (prior 5s rates ÷ 5).

Model ID Credits / sec Notes
ByteDance/Seedance-1.0-lite 80 Seedance 1.0 Lite

Voice models