Supported Models

Fleek provides optimized inference for leading open-source AI models across text, image, video, and 3D generation. All models run at $0.0025 per GPU-second with no per-token or per-prediction markup.

Browse all models →


Large Language Models (LLMs)

ModelProviderParametersContextUse Case
DeepSeek R1DeepSeek671B MoE (37B active)256KAdvanced reasoning, math, code
Kimi K2.5Moonshot AI1T MoE (32B active)256KMultimodal, agentic workflows
GLM 4.7Z.ai355B MoE (32B active)200KCode generation (73.8% SWE-bench)
Qwen3-235BAlibaba235B128KMultilingual, general reasoning
Llama 70BMeta70B128KProduction workloads
gpt-oss-120bCommunity120B32KHigh throughput, clean IP

Coding-Optimized

ModelProviderParametersContextHighlights
Qwen3 Coder 480BAlibaba480B MoE256K67% SWE-bench, Apache 2.0
Qwen3 Coder 30B A3BAlibaba30B MoE (3B active)256KLightweight, fast inference

Image Generation

ModelProviderQualityResolutionFeatures
FLUX.2Black Forest LabsPremium2048×2048Multi-reference editing, color control
Z-ImageAlibabaQuality2048×2048DiT architecture, LoRA-friendly
Qwen Image 2512AlibabaQuality2048×2048Prompt enhancement, style transfer
SDXL TurboStability AIFast1024×1024Real-time, 1-step generation

Image Editing

ModelProviderFeatures
Qwen Edit 2511AlibabaInstruction-based editing, style transfer, object removal

Video Generation

ModelProviderDurationResolutionFeatures
LTX-2 19BLightricksUp to 20s4KAudio sync, lip sync, 50fps
Wan MoveAlibabaUp to 15s1080pImage-to-video, motion control
Wan 2.2 14BAlibabaUp to 16s1080pText-to-video, temporal consistency
SeedVRByteDanceUp to 10s4KVideo upscaling, detail enhancement
Seedance 1.5ByteDanceUp to 12s1080pDance generation, beat sync

3D Generation

ModelProviderOutput FormatsFeatures
Hunyuan 3D-2TencentGLB, OBJ, FBXText-to-3D, image-to-3D, game-ready
Trellis 2 4BMicrosoftGLB, USDZ, OBJFast, MIT licensed, Apple support

Model Selection Guide

For reasoning & analysis

DeepSeek R1 — Best-in-class chain-of-thought reasoning

For coding tasks

GLM 4.7 — Highest SWE-bench score (73.8%) → Qwen3 Coder 480B — Handles entire codebases

For multimodal workflows

Kimi K2.5 — Native image/screenshot analysis

For production scale

Llama 70B — Proven reliability, broad support

For image generation

FLUX.2 — Photorealistic quality → SDXL Turbo — Real-time generation

For video

LTX-2 19B — Only model with native audio sync


Coming Soon

We're constantly adding new models. Request a model on Discord or contact us.


Questions?