The cheapest provider for every model.
43 text LLMs + 72 image & video models. Text per 1M tokens (in / out); media normalized to one reference output (16:9 1080p image · 5-second 1080p clip). The green cell is the cheapest route ai-lcr picks first.
Text LLMs · $/1M tokens, input / output
| Model | Official | Best Value | OpenRouter | TokenMart | DeepInfra |
|---|---|---|---|---|---|
DeepSeek V4 Flashopen weights DeepSeek · deepseek/deepseek-v4-flash | $0.14 / $0.28 | TokenMart−37% | $0.1 / $0.2 | $0.09 / $0.18 | $0.1 / $0.2 |
GPT-5 nanoproprietary OpenAI · openai/gpt-5-nano | $0.05 / $0.4 | TokenMart−40% | $0.05 / $0.4 | $0.03 / $0.24 | — |
Gemini 2.5 Flash-Liteproprietary Google · google/gemini-2.5-flash-lite | $0.1 / $0.4 | TokenMart−40% | $0.1 / $0.4 | $0.06 / $0.24 | — |
Qwen3.5 Flashopen weights Alibaba · alibaba/qwen3.5-flash | $0.1 / $0.4 | OpenRouter−35% | $0.07 / $0.26 | $0.07 / $0.28 | — |
GLM-4.5-Airopen weights Z.ai · zai/glm-4.5-air | $0.2 / $1.1 | TokenMart−55% | $0.13 / $0.85 | $0.08 / $0.51 | — |
GPT-5.4 nanoproprietary OpenAI · openai/gpt-5.4-nano | $0.2 / $1.25 | TokenMart−40% | $0.2 / $1.25 | $0.12 / $0.75 | — |
Gemini 3.1 Flash-Liteproprietary Google · google/gemini-3.1-flash-lite | $0.25 / $1.5 | TokenMart−40% | $0.25 / $1.5 | $0.15 / $0.9 | — |
GPT-4.1 miniproprietary OpenAI · openai/gpt-4.1-mini | $0.4 / $1.6 | TokenMart−40% | $0.4 / $1.6 | $0.24 / $0.96 | — |
MiniMax M2.5open weights MiniMax · minimax/minimax-m2.5 | $0.3 / $1.2 | TokenMart−19% | $0.15 / $1.15 | $0.24 / $0.97 | $0.15 / $1.15 |
MiniMax M2.7open weights MiniMax · minimax/minimax-m2.7 | $0.3 / $1.2 | TokenMart−19% | $0.28 / $1.2 | $0.24 / $0.97 | — |
GLM-4.7open weights Z.ai · zai/glm-4.7 | $0.6 / $2.2 | TokenMart−54% | $0.4 / $1.75 | $0.24 / $1.05 | $0.4 / $1.75 |
GLM-4.6open weights Z.ai · zai/glm-4.6 | $0.6 / $2.2 | TokenMart−54% | $0.43 / $1.74 | $0.26 / $1.04 | $0.43 / $1.74 |
DeepSeek V4 Proopen weights DeepSeek · deepseek/deepseek-v4-proOfficial reflects DeepSeek's permanent price (the launch promo became standard on 2026-05-31, 1/4 of original list). | $0.43 / $0.87 | OpenRouter | $0.43 / $0.87 | $0.43 / $0.87 | $1.3 / $2.6 |
GPT-5 miniproprietary OpenAI · openai/gpt-5-mini | $0.25 / $2 | TokenMart−40% | $0.25 / $2 | $0.15 / $1.2 | — |
Qwen3.5 Plusopen weights Alibaba · alibaba/qwen3.5-plus | $0.4 / $2.4 | TokenMart−48% | $0.26 / $1.56 | $0.21 / $1.26 | — |
GLM-5open weights Z.ai · zai/glm-5 | $1 / $3.2 | TokenMart−62% | $0.6 / $2.08 | $0.36 / $1.25 | $0.6 / $2.08 |
Gemini 2.5 Flashproprietary Google · google/gemini-2.5-flash | $0.3 / $2.5 | TokenMart−40% | $0.3 / $2.5 | $0.18 / $1.5 | — |
GLM-4.5open weights Z.ai · zai/glm-4.5 | $0.6 / $2.2 | TokenMart−40% | $0.6 / $2.2 | $0.36 / $1.32 | — |
Gemini 3 Flashproprietary Google · google/gemini-3-flash-preview | $0.5 / $3 | TokenMart−40% | $0.5 / $3 | $0.3 / $1.8 | — |
Kimi K2.5open weights Moonshot · moonshot/kimi-k2.5 | $0.6 / $3 | OpenRouter−36% | $0.4 / $1.9 | $0.42 / $2.1 | $0.45 / $2.25 |
GLM-5.1open weights Z.ai · zai/glm-5.1 | $1.4 / $4.4 | TokenMart−53% | $0.98 / $3.08 | $0.63 / $2.1 | $1.05 / $3.5 |
GLM-5-Turboopen weights Z.ai · zai/glm-5-turbo | $1.2 / $4 | TokenMart−40% | $1.2 / $4 | $0.72 / $2.4 | — |
GPT-5.4 miniproprietary OpenAI · openai/gpt-5.4-mini | $0.75 / $4.5 | TokenMart−40% | $0.75 / $4.5 | $0.45 / $2.7 | — |
Kimi K2.6open weights Moonshot · moonshot/kimi-k2.6 | $0.95 / $4 | TokenMart−31% | $0.68 / $3.42 | $0.63 / $2.8 | $0.75 / $3.5 |
Grok 4.3proprietary xAI · xai/grok-4.3 | $1.25 / $2.5 | TokenMart−1% | $1.25 / $2.5 | $1.2 / $2.5 | — |
Claude Haiku 4.5proprietary Anthropic · anthropic/claude-haiku-4-5-20251001 | $1 / $5 | TokenMart−35% | $1 / $5 | $0.65 / $3.25 | — |
Qwen3.7 Maxopen weights Alibaba · alibaba/qwen3.7-maxNo first-party Alibaba list price found; OpenRouter shown as reference. | — | OpenRouter−50% | $1.25 / $3.75 | $1.75 / $5.25 | $2.5 / $7.5 |
GPT-4.1proprietary OpenAI · openai/gpt-4.1 | $2 / $8 | TokenMart−40% | $2 / $8 | $1.2 / $4.8 | — |
Gemini 3.5 Flashproprietary Google · google/gemini-3.5-flash | $1.5 / $9 | TokenMart−40% | $1.5 / $9 | $0.9 / $5.4 | — |
GPT-5proprietary OpenAI · openai/gpt-5 | $1.25 / $10 | TokenMart−40% | $1.25 / $10 | $0.75 / $6 | — |
GPT-5.1proprietary OpenAI · openai/gpt-5.1 | $1.25 / $10 | TokenMart−40% | $1.25 / $10 | $0.75 / $6 | — |
Gemini 2.5 Proproprietary Google · google/gemini-2.5-pro | $1.25 / $10 | TokenMart−40% | $1.25 / $10 | $0.75 / $6 | — |
Gemini 3 Proproprietary Google · google/gemini-3-pro-previewOfficial = Gemini 3 Pro ≤200k tier. | $2 / $12 | TokenMart−40% | — | $1.2 / $7.2 | — |
Gemini 3.1 Proproprietary Google · google/gemini-3.1-pro-preview | $2 / $12 | TokenMart−40% | $2 / $12 | $1.2 / $7.2 | — |
GPT-5.2proprietary OpenAI · openai/gpt-5.2 | $1.75 / $14 | TokenMart−40% | $1.75 / $14 | $1.05 / $8.4 | — |
GPT-5.2 Codexproprietary OpenAI · openai/gpt-5.2-codex | $1.75 / $14 | TokenMart−40% | $1.75 / $14 | $1.05 / $8.4 | — |
GPT-5.3 Codexproprietary OpenAI · openai/gpt-5.3-codex | $1.75 / $14 | TokenMart−40% | $1.75 / $14 | $1.05 / $8.4 | — |
GPT-5.4proprietary OpenAI · openai/gpt-5.4 | $2.5 / $15 | TokenMart−40% | $2.5 / $15 | $1.5 / $9 | — |
Claude Sonnet 4.6proprietary Anthropic · anthropic/claude-sonnet-4-6 | $3 / $15 | TokenMart−35% | $3 / $15 | $1.95 / $9.75 | — |
Claude Opus 4.6proprietary Anthropic · anthropic/claude-opus-4-6 | $5 / $25 | TokenMart−35% | $5 / $25 | $3.25 / $16.25 | — |
Claude Opus 4.7proprietary Anthropic · anthropic/claude-opus-4-7 | $5 / $25 | TokenMart−35% | $5 / $25 | $3.25 / $16.25 | — |
Claude Opus 4.8proprietary Anthropic · anthropic/claude-opus-4-8 | $5 / $25 | TokenMart−35% | $5 / $25 | $3.25 / $16.25 | — |
GPT-5.5proprietary OpenAI · openai/gpt-5.5 | $5 / $30 | TokenMart−40% | $5 / $30 | $3 / $18 | — |
Image & Video · normalized ¢ per reference output
| Model | Official | Best Value | Kunavo | fal | Runware | WaveSpeed | Replicate |
|---|---|---|---|---|---|---|---|
FLUX.2 Klein 4Bopen weights Black Forest Labs · bfl/flux-2-klein-4b | 1.4¢ | Runware−96% | — | — | 0.06¢ | 0.8¢ | 0.1¢ |
Flux Schnellopen weights Black Forest Labs · bfl/flux-schnell | — | Runware−57% | — | 0.3¢ | 0.13¢ | 0.3¢ | 0.3¢ |
Z-Image Turboopen weights Alibaba · alibaba/z-image-turbo | 1.5¢ | Runware−91% | — | — | 0.13¢ | 0.5¢ | 0.5¢ |
Stable Diffusion 3open weights Stability AI · stability/stable-diffusion-3 | 6.5¢ | Runware−97% | — | — | 0.19¢ | 3¢ | 3.5¢ |
RealVisXL V5.0open weights Civitai · civitai/realvisxl-v5 | — | Runware | — | — | 0.19¢ | — | — |
DreamShaper XLopen weights Civitai · civitai/dreamshaper-xl | — | Runware | — | — | 0.19¢ | — | — |
HiDream-I1 Fastopen weights HiDream · hidream/hidream-i1-fast | — | Runware−48% | — | — | 0.26¢ | — | 0.5¢ |
Fast SDXLopen weights Stability AI · stability/fast-sdxl | 0.9¢ | fal−67% | — | 0.3¢ | — | — | — |
Qwen-Imageopen weights Alibaba · alibaba/qwen-image | 3.5¢ | Runware−89% | — | — | 0.38¢ | 2¢ | 2.5¢ |
Qwen-Image-Edit Plusopen weightsedit Alibaba · alibaba/qwen-image-edit | 4.5¢ | Runware−91% | — | — | 0.4¢ | 2¢ | 3¢ |
Z-Imageopen weights Alibaba · alibaba/z-image | — | Runware | — | — | 0.45¢ | — | — |
HiDream-I1 Devopen weights HiDream · hidream/hidream-i1-dev | — | Runware−63% | — | — | 0.45¢ | 1.2¢ | — |
BiRefNet Background Removalopen weightsbg-removal fal · fal/birefnet | — | fal | — | 0.5¢ | — | — | — |
FLUX.2 [dev]open weights Black Forest Labs · bfl/flux-2-dev | — | Runware−36% | — | — | 0.9¢ | 1.2¢ | 1.4¢ |
FLUX.1 Krea [dev]open weights Black Forest Labs · bfl/flux-1-krea-dev | — | Runware−63% | — | — | 0.92¢ | 2¢ | 2.5¢ |
FLUX.2 Klein 9Bopen weights Black Forest Labs · bfl/flux-2-klein-9b | 1.5¢ | WaveSpeed−33% | — | — | 2¢ | 1¢ | 1.5¢ |
Flux Devopen weights Black Forest Labs · bfl/flux-dev | — | WaveSpeed−52% | — | 2.5¢ | 2.5¢ | 1.2¢ | 2.5¢ |
Bria RMBG 2.0proprietarybg-removal Bria · bria/rmbg-2 | 1.8¢ | fal | — | 1.8¢ | — | 1.8¢ | — |
Grok Imagine (Quality)proprietary xAI · xai/grok-image-qualityCross-tier — this row is the Quality tier (Official $0.07/img). WaveSpeed/Replicate quote the standard Grok image tier (~$0.02), not Quality, so their lower price isn't like-for-like. | 7¢ | Replicate−71% | — | — | 5¢ | 2.2¢ | 2¢ |
Nano Bananaproprietary Google · google/nano-bananaGemini Flash Image v1 — Kunavo only (fal/runware carry v2/pro) | 3.9¢ | Kunavo−30% | 2.73¢ | — | — | 3.8¢ | 3.9¢ |
Kling IMAGE 3.0proprietary Kuaishou · kuaishou/kling-image-3 | 2.8¢ | Runware | — | — | 2.8¢ | 2.8¢ | — |
Kling IMAGE O3proprietary Kuaishou · kuaishou/kling-image-o3 | 2.8¢ | Runware | — | — | 2.8¢ | 2.8¢ | — |
FLUX.2 [pro]proprietary Black Forest Labs · bfl/flux-2-pro | 3¢ | Runware | — | — | 3¢ | 3¢ | 3¢ |
Wan2.7 Image Proproprietary Alibaba · alibaba/wan-2-7-image-proReplicate lists $0.03/img vs $0.075 at Official/Runware/WaveSpeed — a 2.5× gap worth verifying (likely a smaller default resolution). | 7.5¢ | Replicate−60% | — | — | 7.5¢ | 7.5¢ | 3¢ |
Seedream 4proprietary ByteDance · bytedance/seedream-4-5 | 3¢ | fal | — | 3¢ | — | 4¢ | 4¢ |
GPT Image 1.5proprietary OpenAI · openai/gpt-image-1-5 | 3.4¢ | Runware | — | — | 3.39¢ | 4¢ | 13.6¢ |
Seedream 5.0 Liteproprietary ByteDance · bytedance/seedream-5-lite | 3.5¢ | Runware | — | — | 3.5¢ | 3.5¢ | 3.5¢ |
Flux 1.1 Proproprietary Black Forest Labs · bfl/flux-1.1-pro | 4¢ | Runware | — | 4¢ | 4¢ | 4¢ | 4¢ |
Flux Kontext Proproprietaryedit Black Forest Labs · bfl/flux-kontext-pro | 4¢ | fal | — | 4¢ | — | 4¢ | 4¢ |
Recraft V3proprietary Recraft · recraft/v3 | 4¢ | fal | — | 4¢ | — | 4¢ | 4¢ |
Recraft V4.1proprietary Recraft · recraft/v4-1 | 25¢ | Runware−84% | — | — | 4¢ | 4¢ | — |
Nano Banana 2proprietary Google · google/nano-banana-2Runware 1K tier (2K 10.3¢, 4K 15.3¢) | 6.7¢ | Kunavo−30% | 4.69¢ | 8¢ | 6.9¢ | 7¢ | 6.7¢ |
Runway Gen-4 Imageproprietary Runway · runway/gen-4-imageReplicate's $0.05 is 720p; Official/fal/Runware/WaveSpeed are 1080p ($0.08). At equal 1080p, Replicate is ~$0.08 too. | 8¢ | Replicate−38% | — | — | 8¢ | 8¢ | 5¢ |
Clarity Upscaleropen weightsupscale fal · fal/clarity-upscaler | — | fal | — | 5¢ | — | — | — |
Imagen 4 Ultraproprietary Google · google/imagen-4-ultra | 6¢ | WaveSpeed−3% | — | 6¢ | 6¢ | 5.8¢ | 6¢ |
FLUX.2 [flex]proprietary Black Forest Labs · bfl/flux-2-flex | 6¢ | Runware | — | — | 6¢ | 6¢ | 6¢ |
GPT Image 2proprietary OpenAI · openai/gpt-image-2Quality tiers differ: Official/WaveSpeed are medium-1K; Replicate bills per token at auto/high. Match quality before comparing. | 5.3¢ | WaveSpeed | 6.33¢ | — | 9.38¢ | 6¢ | 12.8¢ |
Ideogram V3 Balancedproprietary Ideogram · ideogram/v3-balanced | 6¢ | Runware | — | 6¢ | 6¢ | 6¢ | 6¢ |
Nano Banana Proproprietary Google · google/nano-banana-pro | 13.4¢ | Kunavo−50% | 6.7¢ | 8¢ | — | 14¢ | 15¢ |
Flux Kontext Maxproprietaryedit Black Forest Labs · bfl/flux-kontext-max | 8¢ | fal | — | 8¢ | — | 8¢ | 8¢ |
WAN 2.2 (T2V)open weights Alibaba · alibaba/wan-2-2-t2vReplicate's $0.10 is the distilled wan-2.2-t2v-fast variant — not the full model WaveSpeed/Runware price at 720p. | 50¢ | Replicate−80% | — | — | 28.85¢ | 30¢ | 10¢ |
HunyuanVideo 1.5open weights Tencent · tencent/hunyuan-video-1-5 | — | Runware−42% | — | — | 11.65¢ | 20¢ | — |
GPT Image 2 (High)proprietary OpenAI · openai/gpt-image-2-highReplicate bills gpt-image-2 by token at auto quality (~$0.128); fal/WaveSpeed/Official quote the fixed High-1K tier (~$0.21) — Replicate's effective detail can be lower. | 21.1¢ | Replicate−39% | — | 21¢ | — | 22¢ | 12.8¢ |
Veo 3.1 Liteproprietary Google · google/veo-3-1-liteKunavo flat per clip (SKU unverified); fal per second | 40¢ | Kunavo−60% | 16¢ | 40¢ | — | 40¢ | 40¢ |
Seedance Liteproprietary ByteDance · bytedance/seedance-v1-lite | — | fal−60% | — | 18¢ | — | 45¢ | 36¢ |
Luma Ray 2 (I2V)proprietary Luma · luma/ray-2-i2vWaveSpeed bills Ray 2 flat per-clip (~$0.20, 720p); Replicate/fal bill per-second — the flat rate wins at 5s but isn't a per-second comparison. | — | WaveSpeed−90% | — | 200¢ | — | 20¢ | 90¢ |
Hailuo 02 Standardproprietary MiniMax · minimax/hailuo-02-standardOfficial = nearest tier: 768p, 6s (no 5s/1080p SKU). | 27¢ | fal−17% | — | 22.5¢ | — | 23¢ | 27¢ |
LTX-2open weights Lightricks · lightricks/ltx-2 | 30¢ | fal | — | 30¢ | — | 30¢ | 30¢ |
LTX-2 (I2V)open weights Lightricks · lightricks/ltx-2-i2v | 30¢ | fal | — | 30¢ | — | 30¢ | 30¢ |
WAN 2.2 (I2V)open weights Alibaba · alibaba/wan-2-2-i2vWAN 2.2 is 720p-only at every provider (no 1080p tier); per-clip prices still vary a lot by host. | 50¢ | WaveSpeed−40% | — | 40¢ | — | 30¢ | 100¢ |
Veo 3.1proprietary Google · google/veo-3-1Kunavo bills flat per clip (~8s, SKU unverified); fal is veo3.1 per second, 1080p audio-on (20¢/s audio-off) | 200¢ | Kunavo−84% | 32¢ | 200¢ | — | 200¢ | 200¢ |
Kling 2.6 Pro (I2V)proprietary Kuaishou · kuaishou/kling-v26-pro-i2v | 56¢ | fal−38% | — | 35¢ | — | 35¢ | 35¢ |
Kling Motion Controlproprietary Kuaishou · kuaishou/kling-motion-controlPrices span different Kling versions/tiers (v2.6 vs v3, std vs pro) — not a like-for-like motion-control SKU. | 56¢ | Replicate−38% | — | 49¢ | — | 56¢ | 35¢ |
Hailuo 02 Proproprietary MiniMax · minimax/hailuo-02-proOfficial = nearest tier: 1080p, 6s (no 5s SKU). | 53¢ | fal−25% | — | 40¢ | — | 48¢ | 48¢ |
PixVerse v5.5 (I2V)proprietary PixVerse · pixverse/v5-5-i2v | 60¢ | fal−33% | — | 40¢ | — | 160¢ | — |
PixVerse V6proprietary PixVerse · pixverse/v6 | 45¢ | Runware−11% | — | — | 40¢ | 45¢ | 90¢ |
Hailuo 2.3 Proproprietary MiniMax · minimax/hailuo-2-3-proOfficial = nearest tier: 1080p, 6s (no 5s SKU). | 53¢ | fal−8% | — | 49¢ | — | 49¢ | 49¢ |
WAN 2.7proprietary Alibaba · alibaba/wan-2-7-t2vReplicate bills a flat $0.10/sec (resolution-agnostic); the others are $0.15/sec at 1080p. | 75¢ | Replicate−33% | — | — | 75¢ | 75¢ | 50¢ |
Sora 2proprietary OpenAI · openai/sora-2Official is Sora 2 at 720p (base has no 1080p tier). | 50¢ | fal | — | 50¢ | — | 50¢ | 50¢ |
Sora 2 (I2V)proprietary OpenAI · openai/sora-2-i2vOfficial is Sora 2 at 720p (base has no 1080p tier). | 50¢ | fal | — | 50¢ | — | 50¢ | 50¢ |
Kling V3 Proproprietary Kuaishou · kuaishou/kling-v3-pro | 56¢ | fal | — | 56¢ | — | 56¢ | 112¢ |
Kling V3 Pro (I2V)proprietary Kuaishou · kuaishou/kling-v3-pro-i2v | 56¢ | fal | — | 56¢ | — | 56¢ | 112¢ |
Seedance 1.0 Proproprietary ByteDance · bytedance/seedance-v1-pro | 61¢ | WaveSpeed−2% | — | 62¢ | — | 60¢ | 75¢ |
Seedance 1.0 Pro (I2V)proprietary ByteDance · bytedance/seedance-v1-pro-i2v | 61¢ | WaveSpeed−2% | — | 62¢ | — | 60¢ | 75¢ |
WAN 2.5 (I2V)proprietary Alibaba · alibaba/wan-2-5-i2v | 75¢ | fal | — | 75¢ | — | 75¢ | 75¢ |
WAN 2.5 (T2V)proprietary Alibaba · alibaba/wan-2-5-t2v | 75¢ | fal | — | 75¢ | — | 75¢ | 75¢ |
Seedance 2.0 Fastproprietary ByteDance · bytedance/seedance-2-0-fastFast has no 1080p tier: Official & Replicate are 720p; WaveSpeed's $2.50 is a 1080p quote, so it's not comparable. | 60¢ | Replicate | — | 120.95¢ | — | 250¢ | 75¢ |
Seedance 2.0 Fast (I2V)proprietary ByteDance · bytedance/seedance-2-0-fast-i2vFast has no 1080p tier: Official & Replicate are 720p; WaveSpeed's $2.50 is a 1080p quote, so it's not comparable. | 60¢ | Replicate | — | 120.95¢ | — | 250¢ | 75¢ |
Seedance 2.0proprietary ByteDance · bytedance/seedance-2-0 | 187¢ | Runware−35% | — | — | 120.95¢ | 300¢ | 225¢ |
Kling 2.1 Masterproprietary Kuaishou · kuaishou/kling-v21-master | 56¢ | WaveSpeed | — | 140¢ | — | 130¢ | 140¢ |
Kling 2.1 Master (I2V)proprietary Kuaishou · kuaishou/kling-v21-master-i2v | 56¢ | WaveSpeed | — | 140¢ | — | 130¢ | 140¢ |
Veo 3 Qualityproprietary Google · google/veo-3-qualityKunavo only — flat per clip (SKU unverified) | 200¢ | Kunavo−4% | 192¢ | — | — | 320¢ | 200¢ |
Notes. Text rates are per 1M tokens (input / output). Officialis the model maker's own first-party list price (OpenAI, Anthropic, Google, Z.ai, DeepSeek, Moonshot, etc.); OpenRouter, TokenMart and DeepInfra are pulled live from each provider's /v1/models. The green cell + Best Valuecolumn mark the cheapest buyable route and its discount versus Official. A listed price ≠ a working route — some discount upstreams aren't provisioned and 502 in practice, so re-probe before routing; the live status page tracks which are actually up. Image & video models are normalized to 16:9 1080p image · 5-second 1080p clip instead (Kunavo bills a flat per-call fee while fal/Runware bill per-second); their Officialcolumn is the maker's first-party API rate at the closest standard tier (open-weight models with no first-party API show “—”), so it's a directional reference, not a SKU-exact match. Same model, different SKU:aggregators (fal, Runware, WaveSpeed, Replicate) sometimes quote a different resolution, duration, quality tier, or a distilled “fast” variant — so a cheaper cell isn't always the same output. Rows where the basis doesn't line up carry an italic note spelling out the difference; trust those over the raw number. License tags are best-effort (open = downloadable weights, proprietary = API-only).