Catalog
Models
Every model the GraphiLink tools call: who made it, where it works, what it costs.
Showing 123 of 123
| Tools | |||||
|---|---|---|---|---|---|
gpt-6-astra | OpenAI | Textreads images | 922Koutput up to 128K | $10.00 in / $50.00 out / cached $1.00 / 1M tokover 272K: $20.00 in / $75.00 out | |
gpt-6-sol | OpenAI | Textreads images | 922Koutput up to 128K | $2.00 in / $10.00 out / cached $0.20 / 1M tokover 272K: $4.00 in / $15.00 out | |
gpt-6-luna | OpenAI | Textreads images | 922Koutput up to 128K | $0.10 in / $0.50 out / cached $0.01 / 1M tokover 272K: $0.20 in / $0.75 out | |
gpt-5.6-sol | OpenAI | Textreads images | 922Koutput up to 128K | $5.00 in / $30.00 out / cached $0.50 / 1M tokover 272K: $10.00 in / $45.00 out | |
gpt-5.6-terra | OpenAI | Textreads images | 922Koutput up to 128K | $2.00 in / $12.00 out / cached $0.20 / 1M tokover 272K: $4.00 in / $18.00 out | |
gpt-5.6-luna | OpenAI | Textreads images | 922Koutput up to 128K | $0.20 in / $1.20 out / cached $0.02 / 1M tokover 272K: $0.40 in / $1.80 out | |
gpt-5.5 | OpenAI | Textreads images | 922Koutput up to 128K | $5.00 in / $30.00 out / cached $0.50 / 1M tokover 272K: $10.00 in / $45.00 out | |
gpt-5.4 | OpenAI | Textreads images | 922Koutput up to 128K | $2.50 in / $15.00 out / cached $0.25 / 1M tokover 272K: $5.00 in / $22.50 out | |
gpt-5.4-mini | OpenAI | Textreads images | 272Koutput up to 128K | $0.75 in / $4.50 out / cached $0.08 / 1M tok | |
gpt-5.4-nano | OpenAI | Textreads images | 272Koutput up to 128K | $0.20 in / $1.25 out / cached $0.02 / 1M tok | |
gpt-5.2 | OpenAI | Textreads images | 272Koutput up to 128K | $1.75 in / $14.00 out / cached $0.18 / 1M tok | |
gpt-5-mini | OpenAI | Textreads images | 272Koutput up to 128K | $0.25 in / $2.00 out / cached $0.03 / 1M tok | |
gpt-5-nano | OpenAI | Textreads images | 272Koutput up to 128K | $0.05 in / $0.40 out / cached <$0.01 / 1M tok | |
gpt-4.1 | OpenAI | Textreads images | 1.05Moutput up to 33K | $2.00 in / $8.00 out / cached $0.50 / 1M tok | |
gpt-4o-mini | OpenAI | Textreads images | 128Koutput up to 16K | $0.15 in / $0.60 out / cached $0.08 / 1M tok | |
anthropic/claude-fable-5.1 | Anthropic | Textreads images | 1Moutput up to 128K | $10.00 in / $50.00 out / cached $0.25 / 1M tok | |
anthropic/claude-opus-5.5 | Anthropic | Textreads images | 1Moutput up to 128K | $4.00 in / $20.00 out / cached $0.20 / 1M tok | |
anthropic/claude-opus-5 | Anthropic | Textreads images | 1Moutput up to 128K | $5.00 in / $25.00 out / cached $0.50 / 1M tok | |
anthropic/claude-opus-4.8 | Anthropic | Textreads images | 1Moutput up to 128K | $5.00 in / $25.00 out / cached $0.50 / 1M tok | |
anthropic/claude-sonnet-4.6 | Anthropic | Textreads images | 1Moutput up to 128K | $3.00 in / $15.00 out / cached $0.30 / 1M tok | |
anthropic/claude-haiku-4.5 | Anthropic | Textreads images | 200Koutput up to 64K | $1.00 in / $5.00 out / cached $0.10 / 1M tok | |
x-ai/grok-4.5 | xAI | Textreads images | 500Koutput up to 450K | $2.00 in / $6.00 out / cached $0.30 / 1M tokover 200K: $4.00 in / $12.00 out | |
google/gemini-3.7-flash | Textreads images | 1.05Moutput up to 66K | $0.75 in / $3.75 out / cached $0.08 / 1M tok | ||
qwen/qwen3.7-max | Qwen | Text | 1Moutput up to 131K | $1.48 in / $4.43 out / cached $0.30 / 1M tok | |
qwen/qwen3.6-plus | Qwen | Textreads images | 1Moutput up to 66K | $0.33 in / $1.95 out / 1M tokover 256K: $1.30 in / $3.90 out | |
qwen/qwen3.8-27b | Qwen | Textreads images | 1Moutput up to 131K | $0.42 in / $3.00 out / cached $0.09 / 1M tok | |
z-ai/glm-5.3 | Z.ai | Text | 1.31Moutput up to 944K | $1.40 in / $4.40 out / cached $0.26 / 1M tok | |
minimax/minimax-m2.5 | MiniMax | Text | 205Koutput up to 128K | $0.27 in / $1.08 out / cached $0.03 / 1M tok | |
deepseek/deepseek-v4-flash | DeepSeek | Text | 1.05Moutput up to 384K | $0.14 in / $0.28 out / cached $0.03 / 1M tok | |
gpt-5.5-pro | OpenAI | Textreads images | — | $30.00 in / $180.00 out / 1M tokover 272K: $60.00 in / $270.00 out | |
GPT Image 2gpt-image-2higher quality | OpenAI | Images | — | $0.21 / image (high·1024) | |
GPT Image 2.5 Sunburstgpt-image-2.5-sunburstmore precise edits, transparent background | OpenAI | Images | — | $0.05 / image (high·1024) | |
GPT Image 2.5 Flaregpt-image-2.5-flarethe fast 2.5, transparent background | OpenAI | Images | — | $0.05 / image (high·1024) | |
Nano Banana Pro (Gemini 3 Pro)google/gemini-3-pro-imageGoogle's top editor: 1K-4K, up to 8 references | Images | — | $0.14 / image | ||
Nano Banana (Gemini 2.5 Flash)google/gemini-2.5-flash-imagefast generations and edits, up to 3 references | Images | — | $0.04 / image | ||
Nano Banana 2 Lite (Gemini 3.1)google/gemini-3.1-flash-lite-imagecheapest Gemini, formats up to banner strips | Images | — | billed by tokens | ||
MAI Image 2.5 Promicrosoft/mai-image-2.5-proMicrosoft photorealism, 1 reference | Microsoft | Images | — | billed by tokens | |
Recraft V4.1recraft/recraft-v4.1high aesthetics: photorealism, gradients, 3D renders; 1 reference | Recraft | Images | — | $0.04 / image | |
Recraft V4.1 Prorecraft/recraft-v4.1-prothe same at 2K | Recraft | Images | — | $0.21 / image | |
Recraft V4.1 Utilityrecraft/recraft-v4.1-utilityrestrained: product shots, mockups, icons | Recraft | Images | — | $0.04 / image | |
Recraft V4.1 Utility Prorecraft/recraft-v4.1-utility-proUtility at 2K | Recraft | Images | — | $0.21 / image | |
HunyuanImage 3.0hunyuan-image/v3/text-to-imageTencent 80B MoE: long prompts, in-frame text | Tencent | Images | — | $0.10 / image | |
HunyuanImage 3.0 Instructhunyuan-image/v3/instruct/text-to-imagereasons before rendering and expands the prompt; edits with up to 3 references | Tencent | Images | — | $0.09 / image | |
Pixelcut Background Removerpixelcut/background-removalThe cleanest cut-out of the probe: hair strands, spokes, dandelion filaments, not one failure across 24 pictures. About 13 seconds. | Pixelcut | Images | — | $0.016 / image | |
Bria RMBG 2.0bria/background/removeSteady with no failures, hair is excellent, the finest filaments get lost. Trained on licensed data. About 9 seconds. | Bria | Images | — | $0.018 / image | |
BiRefNet v2birefnet/v2Open model billed by GPU seconds: several times cheaper than the rest and the fastest. Five variants of one endpoint — the model parameter. | BiRefNet | Images | — | $0.003 / image | |
FeyNobgfeynobgQuality close to the best for pennies. Billed by the picture's megapixels, about 8 seconds. | Feyn | Images | — | $0.0017 / image | |
Ideogram Remove Backgroundideogram/remove-backgroundBlurs edges and loses thin structures (spokes, filaments), makes glass patchy. About 17 seconds. | Ideogram | Images | — | $0.01 / image | |
Bria Extract Objectbria/extract-objectKeeps only what the prompt names (in English): the one model that removed the dog next to the person. May drop what the person is holding; refuses when the named object is absent. | Bria | Images | — | $0.02 / image | |
Pixelcut Product Photopixelcut/product-photoCuts the item out and recomposes the frame: centres and rescales it. Transparent backdrop, shadow and margins off. About 15 seconds. | Pixelcut | Images | — | $0.024 / image | |
Recraft Crisp Upscalerecraft/upscale/crispFidelity on a par with the best for the lowest price, about 14 seconds. Always 4×, no knobs. Drops transparency. Has input moderation: an explicit picture gets refused. | Recraft | Images | — | $0.004 / image | |
SeedVR2seedvr/upscale/imageThe sharpest of the faithful ones: best on anime and small-print scans, oversharpens high-contrast lettering. About 15 seconds, no moderation. Drops transparency. | ByteDance | Images | — | $0.005 / image | |
Crystal Upscalerclarityai/crystal-upscalerBest on dense small text and anime, the sharpness closest to the original. About 20 seconds. Drops transparency. | Clarity AI | Images | — | $0.064 / image | |
Topaz Wonder 3.5topaz/upscale/image/generativeThe Topaz generative mode that does not invent: best on feathers, watch movements and old prints. About 30 seconds, up to two minutes. Drops transparency. | Topaz Labs | Images | — | $0.08 / image | |
Topaz Transparenttopaz/upscale/image/transparentThe one of the five that keeps the alpha channel: cut-out hair and a dandelion came back clean. Always 4×, about 90 seconds. For layers straight out of "Remove background". | Topaz Labs | Images | — | $0.08 / image | |
Veo 3.1google/veo-3.1Google Veo via OpenRouter: 1080p with native audio, 4/6/8 s. | Video | — | $0.4 / s | ||
Seedance 2.0bytedance/seedance-2.0Seedance via OpenRouter: 4-15 s, character references, first/last frame, audio. | ByteDance | Video | — | billed by tokens | |
HappyHorse 1.1alibaba/happyhorse-1.1Alibaba HappyHorse via OpenRouter: up to 1080p, 3-15 s, strong image-driven animation. | Alibaba | Video | — | $0.0988 / s | |
Grok Imagine Videox-ai/grok-imagine-videoxAI Grok Imagine via OpenRouter: cheap 720p clips, 1-15 s, quick turnaround. | xAI | Video | — | $0.05 / s | |
Kling v3.0 Prokwaivgi/kling-v3.0-proKling v3.0 Pro via OpenRouter: cinematic 1080p with optional audio, 3-15 s. | Kuaishou | Video | — | from $0.168 / s | |
H3 Max Directorminimax/h3-max-director | MiniMax | Video | — | $0.02 / s | |
Hunyuan3D 2.1hunyuan3d-2.1Our own RunPod pod: one image, a textured GLB, seed and quality presets. Included in the plan. | Tencent | 3DOwn GPU | — | own GPU | |
Hunyuan 3D Pro 3.1hunyuan3d-pro-3.1Follows the prompt most closely, about 500k faces, 20–35 MB file, ~2.5 min. Prompt or image. | Tencent | 3D | — | $0.50 / model | |
Hunyuan 3D Pro 3.0hunyuan3d-pro-3.0Same mesh as 3.1, plus the LowPoly and Sketch modes (a line drawing as input). | Tencent | 3D | — | $0.50 / model | |
Hunyuan 3D Rapidhunyuan3d-rapidFaster and cheaper: about 50k faces, coarser texture, ~1.5 min. A prompt up to 200 characters or an image. | Tencent | 3D | — | $0.30 / model | |
Meshy 7meshy/v7/text-to-3dA prompt of up to 600 characters into a GLB and an FBX with their textures embedded, ~3.5 min (~1.7 min untextured). Polygon count, mesh type, quads or triangles, symmetry and pose are yours to set. To build from a photograph, the same Meshy 7 waits in the picture tab. | Meshy | 3D | — | $1.20 / model | |
Meshy 7meshy/v7/image-to-3dOne to four photographs of the same object into a GLB and an FBX with their textures embedded, ~2.3 min (~1.3 min untextured). A separate picture can steer the colouring. Takes no prompt: the shape comes from the shot. | Meshy | 3D | — | $1.20 / model | |
Meshy 7.1meshy/v7.1/text-to-3dThe same Meshy as 7 with geometry resolution up to 4K: a prompt of up to 600 characters into a GLB and an FBX with their textures embedded. 4K is finer in small details and costs 40 cents more, 2K 20 cents. | Meshy | 3D | — | $1.20 / model | |
Meshy 7.1meshy/v7.1/image-to-3dThe same Meshy as 7 with geometry resolution up to 4K: one to four photographs of an object into a GLB and an FBX with their textures embedded. 4K is for a single shot and costs 40 cents more, 2K 20 cents. One shot at 4K with PBR took ~7 min. | Meshy | 3D | — | $1.20 / model | |
TRELLIS.2trellis-2One photograph of an object, or several from different sides, into a GLB with PBR maps (colour, metallic and roughness). Faster and cheaper than Meshy: 512p ~1 min, 1024p ~1.5 min, 1536p ~2 min. Vertex count and texture size are yours to set. Takes no prompt. | Microsoft | 3D | — | $0.30 / model | |
Rodin Gen-2.5hyper3d/rodin/v2.5/text-to-3dA prompt into a GLB with PBR or baked lighting, ~2.5 min. The mesh is set up front: 4K to 200K quads or 2K to 2M triangles, no separate retopology. HighPack: 4K textures (+$0.80). | Hyper3D | 3D | — | $0.40 / model | |
Rodin Gen-2.5hyper3d/rodin/v2.5One to five photographs of an object into a GLB with PBR or baked lighting, ~2 min. The mesh is set up front: 4K to 200K quads or 2K to 2M triangles. HighPack: 4K textures (+$0.80). | Hyper3D | 3D | — | $0.40 / model | |
Rodin Gen-2.5 Fasthyper3d/rodin/v2.5/text-to-3d/fastThe quick variant of Rodin for drafts: a prompt into a light GLB of up to 20K polygons, $0.10. Lower quality than the main Rodin. | Hyper3D | 3D | — | $0.10 / model | |
Rodin Gen-2.5 Fasthyper3d/rodin/v2.5/fastThe quick variant of Rodin for drafts: one to five photographs into a light GLB of up to 20K polygons, 24 s measured, $0.10. Lower quality than the main Rodin. | Hyper3D | 3D | — | $0.10 / model | |
Tripo H3.1tripo3d/h3.1/text-to-3dA prompt into a dense, detailed model with PBR, ~4.5 min. Without a face limit it comes out at about 1.5M triangles and a ~45 MB GLB, so a limit is worth setting. Quads arrive as an FBX. From $0.10 untextured to $0.55 with HD texture, detailed geometry and quads. | Tripo | 3D | — | $0.20 / model | |
Tripo H3.1tripo3d/h3.1/image-to-3dOne photograph of an object, or two to four views (front, left, back, right), into a dense, detailed model with PBR, ~5 min. Quads arrive as an FBX. From $0.20 untextured to $0.65 with HD texture, detailed geometry and quads. | Tripo | 3D | — | $0.30 / model | |
Tripo P2tripo3d/p2/text-to-3dA prompt into a game-ready low-poly model with a clean mesh: 500 to 50K triangles or up to 25K quads (an FBX, no surcharge), PBR and four texture levels. $1.00 untextured, $1.10–1.30 textured. | Tripo | 3D | — | $1.10 / model | |
Tripo P2tripo3d/p2/image-to-3dOne photograph of an object into a game-ready low-poly model with a clean mesh: 500 to 50K triangles or up to 25K quads (an FBX, no surcharge), PBR and four texture levels. $1.00 untextured, $1.10–1.30 textured. | Tripo | 3D | — | $1.10 / model | |
Tripo P1tripo3d/p1/text-to-3dA prompt into a light low-poly model with a clean mesh of up to 20,000 faces (4,447 triangles and a 3 MB GLB with PBR maps measured), ~2 min. For background props and mobile scenes. | Tripo | 3D | — | $0.50 / model | |
Tripo P1tripo3d/p1/image-to-3dOne photograph of an object into a light low-poly model with a clean mesh of up to 20,000 faces, ~2–4 min. For background props and mobile scenes. | Tripo | 3D | — | $0.50 / model | |
Polygen 1.5 · Retopologyhunyuan3d-retopologyTencent smart topology: a high-poly GLB or OBJ of up to 200 MB into a low-poly mesh without textures or UV, 4–6 min. | Tencent | 3D | — | $1.00 / job | |
Tripo Remeshtripo3d/tripo/remeshA clean mesh at a set density: a GLB, OBJ or FBX of up to 150 MB, 500 to 20,000 faces (up to 10,000 quads), the textures baked onto the new mesh. Quads come as an FBX. $0.30, quads +$0.05, ~3.5 min. | Tripo | 3D | — | $0.30 / job | |
Tripo Segmenttripo3d/tripo/segmentSplits a finished model into its semantic parts, each with its own mesh and material. A GLB, OBJ or FBX of up to 150 MB, $0.40, ~2 min. | Tripo | 3D | — | $0.40 / job | |
Hunyuan 3D · Texture Edithunyuan3d-texture-editTencent texture edit of a finished model: an FBX of up to 200 MB as is, a GLB of up to 60 MB through a conversion, from a prompt (PBR optional) or from a reference picture, ~1 min. | Tencent | 3D | — | $0.60 / job | |
TRELLIS.2 · Retexturetrellis-2/retextureA new texture for a finished model from a reference picture: a GLB or OBJ of up to 100 MB as is, no conversion, no prompt. The geometry stays, the UV is laid out anew. $0.20–0.24, about a minute. | Microsoft | 3D | — | $0.24 / job | |
Hunyuan 3D · UV unwraphunyuan3d-uvTencent UV unwrap: an FBX, OBJ or GLB under 30,000 faces into the same mesh with a UV layout — an FBX and an OBJ without textures, ~1 min. | Tencent | 3D | — | $0.20 / job | |
Hunyuan 3D · Auto-rigginghunyuan3d-riggingTencent auto-rigging: a humanoid in a T-pose (an FBX or GLB of up to 60 MB) gets a 22-bone skeleton, optionally with one of 48 preset animations. The result is an FBX, 16–80 s. | Tencent | 3D | — | $0.20 / job | |
HY-Motion 1.0hunyuan-motionA character animation from a description of the motion: up to 12 seconds at 30 frames a second, put on the auto-rigging skeleton. $0.08 a motion, about 15 seconds of work. | Tencent | 3D | — | $0.08 / job | |
Marble 1.1marble-1.1A world from text, a picture, a pano, four views or a video: splats and a collider, 5–7 minutes. | World Labs | 3D | — | $1.26 / generation | |
Marble 1.1 Plusmarble-1.1-plusThe same, but the model widens its coverage on its own and builds a larger world; up to twice the price, about 10 minutes. | World Labs | 3D | — | $2.46 / generation | |
Marble 1.0 Draftmarble-1.0-draftA draft in half a minute: the same panorama, coarse splats without a metric scale — to check a prompt before the full run. | World Labs | 3D | — | $0.18 / generation | |
Lyria 3 Progoogle/lyria-3-pro-previewGoogle Lyria 3 via OpenRouter. Full-length 48kHz stereo songs with vocals from a text prompt. | Music | — | $0.08 / song | ||
Lyria 3 Clipgoogle/lyria-3-clip-previewGoogle Lyria 3 Clip via OpenRouter. Short 48kHz stereo clips/loops with vocals — lighter and cheaper than Pro. | Music | — | $0.04 / song | ||
MiniMax Music 3minimax/music-3MiniMax Music 3 via fal.ai. A full song with vocals: the music description and the lyrics are separate fields and you pick the length. The render takes a few minutes. | MiniMax | Music | — | $0.002 / s | |
GPT Audio 1.5gpt-audio-1.5OpenAI conversational audio, version 1.5: follows instructions more closely and sounds better outside English. Voice phrases and TTS, not music. | OpenAI | Music | — | billed by tokens | |
ACE-Step 1.5acestep-v15-xl-turbo | ACE Studio | MusicOwn GPU | — | $0.02 / song | |
MAI Voice 2 Flashmicrosoft/mai-voice-2-flashMAI-Voice-2 Flash via OpenRouter: fast expressive speech, 18 locales, emotion styles with intensity control. | Microsoft | Speech | — | $15.00 / 1M chars | |
Qwen Audio 3 TTS Flashqwen/qwen-audio-3.0-tts-flashQwen-Audio-3.0 TTS Flash via OpenRouter: multilingual speech, DashScope timbres, accepts an arbitrary voice id. | Qwen | Speech | — | $15.00 / 1M chars | |
Speech 2.8 Turbominimax/speech-2.8-turboMiniMax Speech 2.8 Turbo via OpenRouter: 40 languages, presets plus arbitrary voice ids, speed control. | MiniMax | Speech | — | $60.00 / 1M chars | |
Speech 2.8 HDminimax/speech-2.8-hdMiniMax Speech 2.8 HD via OpenRouter: same voices, top fidelity and expressiveness, pricier than Turbo. | MiniMax | Speech | — | $100.00 / 1M chars | |
Gemini 3.1 Flash TTSgoogle/gemini-3.1-flash-tts-previewGemini TTS via OpenRouter: 70+ languages, delivery controlled by inline tags like [whispers]/[excited] in the text. Outputs 24 kHz WAV. | Speech | — | $40.00 / 1M chars | ||
ElevenLabs SFX v2elevenlabs/sound-effects/v2The quality benchmark for short effects. The only stereo model with a true seamless loop — the measured wrap point held within 0.1–0.4 dB. Length caps at 22 seconds. | ElevenLabs | Sound effects | — | $0.002 / s | |
MMAudio v2mmaudio-v2/text-to-audioThe cheapest model here — drafts and idea sweeps. Returns MONO (measured: 44.1 kHz, 1 channel), so there is nothing to pan. Slower than the rest: 10-second clips took up to 53 seconds. | MMAudio | Sound effects | — | $0.001 / s | |
Sonilo v1.1sonilo/v1.1/text-to-sound-effectsStereo, and the only one that goes into minutes — 180 seconds. The provider claims the output is licensed for commercial use. We ask for mp3; its own default is AAC. | Sonilo | Sound effects | — | $0.0018 / s | |
CassetteAI SFXcassetteai/sound-effects-generatorA flat price per generation whatever the length — the cheapest option for a 30-second ambience, the opposite for a four-second hit. Stereo WAV at 44.1 kHz, so the files are large. | CassetteAI | Sound effects | — | $0.01 / generation | |
Stable Audio 3 small SFXstable-audio-3/small/sfx/text-to-audioThinks in sound-design scenes rather than single hits. Flat price per generation. Cannot be looped: the measured tail fades into silence (−74 dB) and the wrap point tears. | Stability AI | Sound effects | — | $0.0206 / generation | |
Mirelo SFX 1.6mirelo-ai/sfx1.6/text-to-audioHas its own ambience mode: a seamless tile for looping. MONO and roughly five times the price of ElevenLabs, but it takes lengths from 0.1 seconds up to a minute. Bills one second more than asked. | Mirelo AI | Sound effects | — | $0.01 / s | |
Stable Audio 2.5stable-audio-25/text-to-audioThe priciest: a flat $0.2 per generation, so a three-second sound costs what a three-minute one does. Only earns its keep on long output. Stereo WAV. | Stability AI | Sound effects | — | $0.2 / generation | |
Nova-3deepgram/nova-3Deepgram Nova-3: fast multilingual transcription, fine-grained segments for subtitles. | Deepgram | Transcription | — | $0.0043 / min | |
Chirp 3google/chirp-3Google Chirp 3: 24 GA languages plus 77+ in preview, auto punctuation and a denoiser. Short files only: 45 s goes through, a minute does not. | Transcription | — | $0.016 / min | ||
MAI Transcribe 1.5microsoft/mai-transcribe-1.5Microsoft MAI-Transcribe 1.5: accents and noisy audio, long calls and captions. The most accurate on Russian in our measurement — 4.2% word error rate. | Microsoft | Transcription | — | $0.006 / min | |
Grok STT 1.0x-ai/grok-stt-1.0xAI Grok STT: transcription with word and segment timestamps, multichannel audio. | xAI | Transcription | — | $0.001667 / min | |
Voxtral Small 24Bmistralai/voxtral-small-24b-2507-sttMistral Voxtral Small 24B: the largest of the family, transcription and translation. | Mistral | Transcription | — | $0.003 / min | |
Parakeet TDT 0.6B v3nvidia/parakeet-tdt-0.6b-v3NVIDIA Parakeet TDT v3: language detection across every official EU language, punctuation. | NVIDIA | Transcription | — | $0.0015 / min | |
Voxtral Mini Transcribemistralai/voxtral-mini-transcribeMistral Voxtral Mini Transcribe: meetings, voice notes, podcasts. | Mistral | Transcription | — | $0.003 / min | |
Voxtral Mini 3Bmistralai/voxtral-mini-3b-2507Mistral Voxtral Mini 3B: a compact model for transcription and translation. | Mistral | Transcription | — | $0.001 / min | |
Qwen3 ASR 1.7Bqwen/qwen3-asr-1.7bQwen3 ASR 1.7B: 30 languages and 22 Chinese dialects, segment timestamps. Middling on Russian — 16.7% word error rate against 4.2% for the best. | Qwen | Transcription | — | $0.00045 / min | |
Qwen3 ASR 0.6Bqwen/qwen3-asr-0.6bQwen3 ASR 0.6B: the same language range in a compact model, segment timestamps. | Qwen | Transcription | — | $0.0002 / min | |
Nemotron 3.5 ASRnvidia/nemotron-3.5-asr-streaming-multilingual-0.6bNVIDIA Nemotron 3.5 ASR: 40+ languages at the lowest rate on the list. | NVIDIA | Transcription | — | $0.0002 / min | |
Transcribe 1fish-audio/transcribe-1Fish Audio Transcribe 1: automatic language detection, word and segment timestamps. | Fish Audio | Transcription | — | $0.006 / min | |
Jev 1.13typesafe/jev-1.13A decision model: given a state and typed questions (choice, scale, yes/no) it returns an answer with probabilities and confidence, no prose. | TypeSafe | Decisions | 32K | $0.042 in / 1M tok | |
qwen/qwen3-embedding-8b | Qwen | Embeddings | — | $0.01 in / 1M tok | |
cohere/rerank-v3.5 | Cohere | Rerank | — | $0.001 / search |
Prices are in USD — exactly what the balance is charged. Text models are priced per million tokens: input, output and cache; “billed by tokens” means by the tokens actually used.