All Models
GPT Image 2.6 Flare
OpenAI
OpenAI's GPT Image 2.6, default tier. New in 2.6: animated GIF and SVG output, edits guided by a markup image, and a reasoning-effort control. Up to 16 reference images; 1K / 2K / 4K or an exact pixel size. Plain images are priced per image, from $0.01 at 1K low to $0.30 at 4K max; GIF, masks, markup images, transparent backgrounds and edits kept at the source size are billed per token at 10% below OpenAI's official rates (image output $27 / 1M, OpenAI: $30); SVG is a flat $0.0315 per image.
GPT Image 2.6 Sunburst
OpenAI
OpenAI's GPT Image 2.6, highest-fidelity tier: more time per image for fine detail, and edits held tightly to the instruction across many rounds. Same 2.6 abilities as Flare — animated GIF and SVG, markup-image edits, reasoning-effort control, up to 16 references. Plain images are priced per image, from $0.01 at 1K low to $0.47 at 4K max, with 1K and 2K the same as Flare; GIF, masks, markup images, transparent backgrounds and edits kept at the source size are billed per token at Flare's rates, 10% below OpenAI's official ones; SVG is a flat $0.0315 per image.
Doubao Seedream 5.5
Doubao
ByteDance's Seedream 5.5 image model: pixel-level edits that hold across many rounds, native RGBA transparent output, and layer splitting into up to 100 transparent layers. Text-to-image or editing from up to 10 reference images at 1K, 1.5K or 2K. $0.03 per image at 1K / 1.5K and $0.05 at 2K, plus $0.01 per reference image, charged only on success.
Wan 3.0 Video
Alibaba
Alibaba's Wan 3.0 all-in-one video model — one model name covers text-to-video, first/last-frame, and omni-reference (up to 10 images + 5 videos + 5 audio clips, or a document / web page). Native 30-second single-shot output at 30fps with a synced audio track. Pass mode:"prime" for the high-speed tier — measured 58s versus 356s on the same job, for 51–56% more per second depending on resolution. Priced per second: $0.045/s at 480p, $0.09 at 720p, $0.18 at 1080p, with a special price on the standard tier of $0.03/s, $0.06 and $0.12 until Nov 23. Reference images, audio and documents are free; a reference video adds its own duration to the billed seconds.
Kling Motion Control
Kling
AI motion control from Kling: give it a character image and a motion video, and the character performs that dance, gesture or walk. Kling 3.0 ($0.10/s at 720p, $0.15/s at 1080p) and Kling 2.6 ($0.06/s, $0.10/s) through one API, billed on the seconds of your reference video, up to 30 s.
Muse Video
Meta
Muse Video — Meta's video model. Text-to-video, image-to-video from one first frame, or up to four reference images that keep the same person, in 720p landscape (1280×720) or portrait (720×1280), 4 to 20 seconds, with sound generated alongside the picture. Failed generations are not charged; $0.037 per requested second, and 5, 6, 8, 10, 15 or 20-second requests come out exactly that long.
Muse Image
Meta
Muse Image — Meta's image model. Text-to-image, or image editing from up to four reference images, at about 2.5 megapixels, with the aspect ratio picked from your prompt or set by you (1:1, 16:9, 9:16, 3:2, 4:3, 21:9 and more), one image per request. Charged only on success, a flat $0.02 per image.
ElevenLabs Sound Effects v3
ElevenLabs
ElevenLabs' newest sound-effect model (SFX v3): several sounds in one prompt, music and vocals, 0.5–30 s clips and seamless loops.
OmniHuman 2.0
ByteDance
OmniHuman 2.0 — ByteDance's audio-driven talking-avatar model. One portrait image plus a driving audio clip of speech or singing, 4 to 300 seconds long, becomes an MP4 with lip sync, facial expressions and head, hand and body gestures that follow the sound; an optional prompt guides motion and camera. The video runs as long as the audio. $0.091 per second of audio, rounded up to the whole second, with no watermark by default.
GPT-6.1 Terra
OpenAI
OpenAI's GPT-6.1 Terra, released October 2026: a fast, efficient model OpenAI aims at visual reasoning, computer use and natural writing, listed at half the price of GPT-6.1 Sol. $0.50 / $2.50 per 1M tokens on APIMODELS — half the $1 / $5 list — and $0.025 cached. 1.05M context, 128K output, text and image input, knowledge cutoff 1 June 2026, reasoning_effort none to max.
Kling 4.0
Kling
Kling 4.0 — the newest Kling video model from Kuaishou, with sound generated in the same pass on every clip. One model id covers text-to-video, reference-to-video with up to 10 images, 5 videos, 7 named subjects and 3 voices that you point at with @mentions in the prompt, and keyframe-to-video from a first frame, up to 8 middle frames and a last frame. 3-30 seconds at 720p or 1080p. From $0.157/s at 720p and $0.235/s at 1080p; requests with reference video are billed per billed second at $0.094/s or $0.141/s.
GPT-6 Astra
OpenAI
OpenAI's September 2026 flagship — 66% below the $10/$50 list. Through this API you get function calling, built-in web search and code interpreter; the computer-use tool is not available. Its usable gains are long-horizon agents and terminal work: 57.7% on Terminal-Bench 4.0 and 88.0% on SRE-Bench against Sol's 37.3% and 55.9%. 1.05M context, 128K output, text + vision in.
Vidu Q4
Vidu
Vidu Q4 — the newest Vidu video model (Preview) from ShengShu, generating dialogue, sound effects and ambience together with the picture. Animate one first frame, or keep characters and voices consistent with up to 15 reference images and 3 reference audio clips. 3-16 seconds at 540p, 720p, 1080p, 2K or 4K. From $0.042/s (540p); 720p $0.0887/s, 1080p $0.112/s, 4K $0.3639/s, the same price with or without audio.
Seedance 2.5 Lite
ByteDance
Seedance 2.5 Lite — a budget Seedance 2.5 tier for text-to-video and reference-image video: 15 or 30 seconds at 720p for a flat $2.50 per clip, with audio generated in the same pass. Add up to 9 JPG or PNG reference images to keep a character, product or style, in six aspect ratios. No first/last frame, no video or audio reference and no silent output — use seedance-2.5 for those. Some jobs fail on this budget tier; failed jobs are refunded automatically.
Doubao Seedream 5.0 Pro
Doubao
Doubao Seedream 5.0 Pro — flagship tier with interactive editing (coordinates / box-select / arrows / hand-drawn marks) and strong multilingual in-image text rendering. Text-to-image, single- and multi-image (2–10) fusion editing, 1K / 2K single-image output, no watermark.
FLUX 3 Max
Black Forest Labs
FLUX 3 Max — a post-trained, latency-optimized variant of FLUX 3 Video from Black Forest Labs, tuned for stronger prompt adherence and visual quality. One model id covers text-to-video, first/last-frame image-to-video and up to 10 timed keyframes; 5-20 seconds at 720p or 1080p, with synced audio generated by default. In our test a 5-second 720p clip was ready 9 seconds after submission through the API. $0.221/s (720p) and $0.377/s (1080p), with a special price of $0.085/s (720p) and $0.145/s (1080p) until Nov 7.
Fish Audio Drama 3
Fish Audio
Fish Audio Drama 3 text-to-speech for acted dialogue: write stage directions such as [barely holding it together] into the text, wrap words in <whisper> or <emphasis>, and voice up to 16 speakers in one request. 83 languages. $0.065 per 1,000 UTF-8 bytes, with a special price of $0.025 until Nov 12.
Media Downloader API
APIMODELS
Turn a public post link from YouTube, TikTok, Instagram, Facebook, X, Xiaohongshu, Bilibili, Douyin or any of 1000+ sites into direct links for its videos, images and audio, with every available quality. Page through whole channels and profiles, pull YouTube and Bilibili subtitles, or get resumable YouTube downloads up to 1080p. Synchronous JSON; failed requests are not charged. From $0.004 per request.
Grok Voice 3.0
xAI
xAI's speech-to-speech model for realtime voice agents, released 14 October 2026; xAI cites Artificial Analysis ranking it #1 on the Speech-to-Speech Index (86.9). Open one WebSocket and talk to it: 28 built-in voices, 20+ languages with automatic detection, server-side turn detection and function calling. $0.104 per minute of connected session, billed per second.
VEO 3.1 Fast Full HD
VEO 3.1 Fast Full HD (1080p) video generation. 8s fixed duration, 16:9 / 9:16, one image as the first frame or two images for a first-to-last-frame transition. When Full HD capacity is short, a request may be delivered at 720p instead of failing.
VEO 3.1 Fast HD
VEO 3.1 Fast HD (720p) video generation. 8s fixed duration, 16:9 aspect ratio, reference image support.
Real-ESRGAN Upscaler
Real-ESRGAN
Real-ESRGAN AI upscaler — best for illustration, anime and hand-drawn artwork: enlarge to 4x HD, 8K and 10K print quality with crisp lines and clean flats. Optional face enhancement. $0.005 per image.
Kling Video-O1
Kling
Kling Video-O1 (Omni-Video) with MVL multi-modal understanding. Supports text, image, video, and subject references.
Kling V3 Omni
Kling
Kling V3 Omni-Video with extended duration and keep-original-sound support for video editing. Official per-second pricing from $0.088/s.
Kling V2.6
Kling
Kling V2.6 video generation with audio and voice customization support. Supports text-to-video and image-to-video.
Gemini 4 Argon
Google's flagship reasoning model, announced 30 Sep 2026 and on the Gemini API from 13 Oct. Per Google: DeepSWE v1.1 77.9%, LVBench 91.7%, CWE-bench v1 68% (tied first), AutomationBench 51.3% (first). 1M context, up to 65,536 output tokens per response. $1.10 / $5.50 per 1M tokens through 30 Nov, $2.20 / $11.00 from 1 Dec 2026 — 45% below Google's list either way.
MiniMax H3 RH Enhanced
MiniMax
MiniMax H3 RH Enhanced — an enhanced build of the open-source MiniMax H3 video model with three modes behind one model id: text-to-video, first/last-frame image-to-video, and multi-reference video (up to 9 reference images, 3 reference videos, 3 voice references). Drive lip sync with your own audio track and choose how the soundtrack is handled. Billed per second: 480P $0.05/s, 768P $0.07/s, 1080P $0.10/s, 4-15 seconds.
Claude Haiku 5.5
Anthropic
Anthropic's Haiku 5.5, released 7 October 2026, for high-volume work. Per Anthropic: GDPval-AA 1620 (Haiku 4.5: 735) and OSWorld 2.1 72.4%, ahead of GPT-6 Luna on every benchmark both report. 1M context, 128K output, adaptive thinking (default effort medium). $0.09 / $0.45 per 1M tokens on APIMODELS — 10% below the $0.10 / $0.50 list — with cache reads at $0.009; prompts over 100K tokens bill at 5×.
Nano Banana 2.1
Google's Nano Banana 2.1 for image generation and editing — up to 10 reference images, 15 aspect ratios, and 1K/2K/4K output at $0.022 / $0.036 / $0.06 per image.
Claude Sonnet 5.5
Anthropic
Anthropic's Sonnet 5.5, released 28 September 2026: 30%+ faster output than Sonnet 5 and up to 30% less per task, with Terminal-Bench 4.0 at 70.6% against 10.3%, per Anthropic. 1M context, 128K output, adaptive thinking (default effort high), knowledge cutoff June 2026. $1.20 / $6.00 per 1M tokens on APIMODELS — 40% below the $2 / $10 list — with cache reads at $0.12 and cache writes at $1.50.
GPT-6.1 Sol
OpenAI
OpenAI's upgraded Sol, released 29 September 2026: higher scores than GPT-6 Sol on coding and automation benchmarks at the same $2 / $10 list, with cached input halved to $0.10. $1.00 / $5.00 per 1M tokens on APIMODELS — half the list — and $0.05 cached. 1.05M context, 128K output, text and image input, knowledge cutoff 30 April 2026, reasoning_effort low to max.
Eleven v4
ElevenLabs
ElevenLabs' newest and highest-quality voice model: more natural delivery, audio tags and 85 languages. $0.08 per 1K characters, the same as ElevenLabs' list price.
Eleven v4 Turbo
ElevenLabs
The v4 voice at real-time speed (about 100 ms latency) for voice agents and live apps, 85 languages. $0.04 per 1K characters, half the price of v4.
Wan 2.7 Spicy
Alibaba
The premium tier of the Spicy image-to-video line: a first frame plus a prompt returns a 2 to 15 second clip at 720p or 1080p with a generated stereo audio track, or animated to your own audio file. Relaxed moderation, anime-style and Asian-face scenes, free prompt rewriting. Per-second billing: $0.169/s at 720p, $0.26/s at 1080p, so a 5-second 1080p clip is $1.30. No last frame — Wan 2.2 Spicy is the more affordable silent volume engine.
Wan 2.2 Animate
Alibaba
A character image plus a reference video: in animate mode your character performs the motion and expressions from the video; in replace mode the person in the video is swapped for your character while scene, camera and motion stay. Output is as long as the reference, at 480p or 720p, and keeps its audio. Billed per second of the reference video: $0.0742/s at 480p, $0.154/s at 720p — a 5-second dance is $0.371 or $0.77, fixed the moment the task is created.
Doubao Seedream 5.0 Flash
Doubao
Doubao Seedream 5.0 Flash — ByteDance's fast, cost-efficient Seedream tier. Text-to-image and image editing with up to 10 reference images, edits that keep an existing PNG's transparency, and layer decomposition (layer splitting) of one image into a base plus up to 16 named transparent PNG layers with bounding boxes. 1K / 2K output in about 10–15 seconds, $0.03 per output image, no watermark.
Claude Opus 5.5
Anthropic
Anthropic's first Claude 5.5 model, released 22 September 2026: Fable 5.1-level results on most work at 40% lower running cost than Opus 5, per Anthropic. 1M context, 128K output, adaptive thinking always on, knowledge cutoff June 2026. $2.40 / $12.00 per 1M tokens on APIMODELS — 40% below the $4 / $20 list — with cache reads at $0.12.
GPT-6 Sol
OpenAI
OpenAI's GPT-6 workhorse tier, released September 2026 alongside Luna: much of GPT-6 Astra's strength in a faster, more affordable model, at half OpenAI's own $2 / $10 list. 1.05M context, 128K output, text and image input, knowledge cutoff 20 April 2026, reasoning_effort from none to max. $1.00 / $5.00 per 1M tokens, cached input $0.10.
GPT-6 Luna
OpenAI
OpenAI's most efficient GPT-6 model for focused, high-volume tasks, released September 2026 with Sol. 1.05M context, 128K output, text and image input, knowledge cutoff 18 May 2026, reasoning_effort from none to max. $0.09 / $0.45 per 1M tokens on APIMODELS — 10% below the $0.10 / $0.50 list — with cached input at $0.009.
Jev
TypeSafe
TypeSafe's Jev — a System One decision model, not a chat model: send your application state plus typed questions (choice / score / noul) and get answers with probabilities and confidence, evaluated in parallel. $0.05 per 1M input tokens, output free, billed per token with no per-call minimum. Same POST /v1/systemone request and response as TypeSafe, so the official SDK works with only base_url and the key changed.
Fish Audio S2.1 Pro
Fish Audio
Fish Audio S2.1 Pro text-to-speech with twenty free voice clones and 83 languages from one model. Two billing tiers run the same model: $0.03 per 1,000 UTF-8 bytes on the standard tier, or $0.005 on the evaluation tier that trades latency and guarantees for price. Billed per byte, so a Chinese character costs three times a Latin one.
Wan 2.2 Spicy
Alibaba
The volume tier of the Spicy image-to-video line: a first frame plus a prompt returns a silent 5 or 8 second clip at 480p or 720p, with an optional last frame to move toward. Free prompt rewriting, no watermark. Per-second billing: $0.03/s at 480p, $0.06/s at 720p, so a 5-second 480p clip is $0.15. For native audio and up to 15 seconds at 1080p, Wan 2.7 Spicy is the premium tier.
DeepSeek V4.1 Flash
DeepSeek
DeepSeek V4.1 Flash — the newer V4.1 generation of DeepSeek’s fast, high-throughput model. 1M-token context, tool calling, JSON output and streaming; OpenAI-compatible. Billed on DeepSeek’s peak / off-peak schedule.
Qwen3.8 Max
Alibaba
Alibaba Qwen3.8 Max — the newest Qwen flagship. Reasoning-first with graded effort (low / medium / xhigh), 1M-token context, tool calling and streaming; OpenAI-compatible.
Qwen3.8 Flash
Alibaba
Alibaba Qwen3.8 Flash — the most affordable reasoning model in the Qwen3.8 line: same graded thinking effort as Max at a fraction of the price, 1M-token context, tool calling and streaming.
GPT Image 2.5 Flare
OpenAI
OpenAI's GPT Image 2.5 — sharper output, faster generation, and edits that finally hold. Mark the spot you don't like and only that changes; the same character or product stays consistent across many rounds of edits. Native transparent PNG, which GPT Image 2 could not produce. Flare is the default tier: full 2.5 quality at roughly half the latency of GPT Image 2, built for social, e-commerce and high-volume generation.
GPT Image 2.5 Sunburst
OpenAI
OpenAI's GPT Image 2.5 — sharper output, faster generation, and edits that finally hold. Mark the spot you don't like and only that changes; the same character or product stays consistent across many rounds of edits. Native transparent PNG, which GPT Image 2 could not produce. Sunburst is the highest-fidelity tier: it spends more time per image for the finest detail and prompt adherence, for premium commercial visuals.
FlashVSR Video Upscale
FlashVSR
Upscale video to 720p, 1080p, 2K or 4K — restore old footage, sharpen low-bitrate clips, or lift AI-generated video before you edit it. Billed per second of the SOURCE clip, so you know the cost before you submit: $0.02/s at 720p, $0.03/s at 1080p. Pay per use, nothing expires — Topaz charges $0.10 per add-on credit (~$0.067/s of HD) on top of a monthly plan whose credits reset every month.
Omni Flash (Lite)
Omni Flash (Lite) — the same Gemini Omni video model as the Stable tier, served on a more affordable channel, about 63% less. The trade-off is visible: this channel stamps a watermark on its output, we strip it before delivery, and the removal leaves a faint patch in the bottom-right corner. Everything else is identical. If that corner matters to your use case, use Omni Flash (Stable) instead. Text-to-video and image-to-video (up to 3 refs), 720p / 1080p at the same price, 4 / 6 / 8 / 10s, 16:9 or 9:16.
Gemini Omni 1.1 Flash
Gemini Omni 1.1 Flash — Google's next Omni video model, four modes behind one model id: text-to-video, first/last-frame image-to-video, reference-to-video (up to 7 reference images, reusable voices and characters) and video editing from a source clip. 720p / 1080p / 4K, 4 / 6 / 8 / 10s, 16:9 or 9:16, synced audio. Billed per output second — 720p $0.07/s, 1080p $0.10/s, 4K $0.20/s — at least 30% below the official rate; video-input generations are a flat $0.70 (720p/1080p) or $1.05 (4K).
MiniMax H3 Max Turbo
MiniMax
MiniMax H3 Max Turbo — a post-trained, throughput-optimized build of MiniMax H3: clips in seconds instead of minutes, stronger prompt adherence, better aesthetics. Billed per second: 480P $0.06/s, 768P $0.096/s, any whole length from 5 to 15 seconds. Text-to-video in six aspect ratios, or image-to-video from a first frame with an optional last frame. No multi-image references, reference video or audio. Launch pricing.
Claude Fable 5.1
Anthropic
Anthropic's September 2026 flagship — exactly half Anthropic's $10/$50 list price. More than doubles Fable 5 on agentic science (52.6% vs 24.7% on Terminal-Bench-Science) and tops it on every published benchmark. 1M context, native Anthropic Messages API, so Claude Code, the SDK and Cursor work unchanged.
Gemini 3.8 Flash
Google's September 2026 Flash — 40% below Google's own $0.75/$3.75 list. Topped DeepSWE v1.1 at launch and scores 89.4% on Terminal-Bench 2.1; beats Claude Opus 5 on HLE-Verified, Vals Finance Agent v2 and Harvey's Legal Agent. 1M context, built for coding and agents.
MiniMax H3 Lite
MiniMax
MiniMax H3 Lite — the same H3 multimodal video model at a fraction of the price: 480P $0.01/s, 768P $0.02/s, any whole length from 1 to 15 seconds. Text-to-video, up to 9 reference images, up to 3 reference audio clips, or first + last frame — one endpoint routes by what you send. No reference video and no 2K. Best on animation and small-motion shots; on live-action with large movement, faces and limbs deform — treat output as a draft. ⚠️ This is a draft tier: it is reliable for animation, talking heads and small motion, but on live-action with fighting, running, dancing or fast camera moves, faces and limbs deform — no prompt fixes it. Lock composition affordably here, then re-shoot on a higher tier.
GLM-5.3
Zhipu
Zhipu GLM-5.3 — Z.ai's newest reasoning model, built for long-horizon agent work and complex software engineering. 1M-token context, strong function calling, served over the OpenAI-compatible chat-completions endpoint.
Grok Imagine Video 1.5
xAI
xAI's Grok Imagine video model — text-to-video and image-to-video under one model name, with a native synced audio track and watermark-free output. 1-15s clips at 480p / 720p / 1080p, priced per second.
FLUX.2 Klein 4B
Black Forest Labs
FLUX.2 Klein 4B — fast (1-2s) photoreal image generation. Text-to-image and image-to-image with up to 3 reference images, fixed 2MP output, 12 aspect ratios. From $0.006 per image (+$0.0015 per reference).
Digital Human
APIMODELS
Turn a person video + a driving audio clip into a lip-synced talking video. Billed per second of the audio — half the price of image-based lip-sync.
Grok 4.6
xAI
xAI's newest frontier model, built for long-running agents and multi-step work — it matches GPT-5.6 Sol on the Artificial Analysis Intelligence Index. 500K context, text + image input, four reasoning-effort levels, and function calling verified at 12/12 under concurrency with nested schemas. 25% below xAI official ($2 / $6 per 1M).
Grok Imagine Image 2.0
xAI
xAI's Image 2.0, built for images you can ship: instruction-following down to the details, designer-grade typography and layout, and multi-reference editing. Four price tiers from $0.03 — 1K/2K at low or medium quality.
Qwen Image 3.0
Alibaba
Alibaba Qwen Image 3.0 — text-to-image and image editing in one model. Up to 4.5k-token prompts, dense in-image text layout, 10px small-text rendering, 12 languages and 20+ fonts. 1K/2K, flat $0.035 per image; each reference image +$0.004.
Qwen Image 3.0 Pro
Alibaba
Alibaba Qwen Image 3.0 Pro — the higher-fidelity tier. Up to 4.5k-token prompts, dense in-image typesetting for posters / storyboards / menus, 10px small text, micro-expressions, pores and hair strands near photographic realism, 12 languages and 20+ fonts. 1K $0.037 / 2K $0.075; each reference image +$0.004.
Seedance 2.5
ByteDance
ByteDance Seedance 2.5 — the long-form and editing tier: 4-30s in a single pass, up to 50 reference assets (30 images + 10 videos + 10 audios), plus video editing and video extension. 480p / 720p (no 1080p — that stays on Seedance 2.0), mp4 or mov, native synced audio and real-person video via the asset library. From $0.120/s.
GPT Image 2
OpenAI
OpenAI gpt-image-2. Text-to-image and multi-image editing (up to 16 reference images), aspect-ratio control, native 1K / 2K / 4K — $0.025 / $0.03 / $0.05 per image.
GPT Image 2 All
OpenAI
OpenAI gpt-image-2, OpenAI-compatible sync API — drop-in for Codex / Cursor / the OpenAI SDK. Full 1K / 2K / 4K × low / medium / high quality, text-to-image and multi-image editing, priced per resolution × quality from $0.01.
Gemini 3 Pro Image (Pro)
Premium image generation powered by Gemini 3 Pro. 99% success rate. Best quality and reliability. $0.068 per image at 1K/2K, $0.13 at 4K.
Nanobananapro-gemini
Gemini 3 Pro Image via a budget channel. Professional asset creation with advanced reasoning and high-fidelity text rendering.
gemini-3-pro-image
Google's gemini-3-pro-image (GA release). Top-quality, high-fidelity image generation and editing with advanced reasoning. Priced by resolution: 1K/2K $0.08 (40% less than official), 4K $0.13 (46% less than official).
Nanobanana2
Fast image generation powered by Gemini 3.1 Flash. Supports text-to-image and image editing — 1K/2K $0.05, 4K $0.08 per image.
Nanobanana2-gemini
Gemini 3.1 Flash Image via a budget channel. High-performance image generation optimized for speed and high-volume use.
Nanobanana-2-lite
The most affordable Gemini 3.1 Flash Image tier — flat $0.025 per image, no resolution tiers. Text-to-image and editing (up to 10 reference images), 1K output only.
Gemini 3.6 Flash
Same input price as 3.5 Flash with more affordable output, for agentic and long-horizon work where the answer, not the prompt, dominates the bill. For the lowest bill outright, see 3.8 Flash.
gemini-3.1-flash-image
Google's gemini-3.1-flash-image (GA). 2K images cost $0.06 versus Google's $0.101 — 40% less — and 1K is the same $0.06, so 2K is a free upgrade here. 512 $0.04, 4K $0.10. Half of requests return within 14s.
Gemini 2.5 Flash Image
Google's Gemini 2.5 Flash Image at a flat $0.02 per image — the most affordable way to generate or edit an image here. Text-to-image and reference-image editing, one API key, no Google Cloud setup.
Claude Opus 5
Anthropic
Anthropic's newest Opus — a step change on deep reasoning, agentic coding and long-horizon work, at 40% off official pricing. 1M context, adaptive thinking on by default, native tool use. Works with Claude Code, Cursor and the OpenAI SDK alike.
Doubao Seedream 5.0 Lite (Official)
Doubao
Doubao Seedream 5.0 Lite via ByteDance Volcano Ark official API. Unified text-to-image and image-to-image (pass image for I2I, omit for T2I). 2K / 4K output, no watermark, PNG.
Doubao Seedream 4.5
Doubao
High quality Doubao Seedream 4.5 image generation. Supports text-to-image and image editing with 2K/4K resolution.
Kling V3 Image
Kling
Kling V3 image generation. Text-to-image and single-reference image-to-image, 1K/2K resolution. $0.05 per image.
Kling V3 Omni
Kling
Kling V3 Omni image generation. Multi-image reference & fusion, element consistency, single/series output, 1K/2K/4K — 1K/2K $0.05, 4K $0.10 per image.
Kling Omni-Image
Kling
AI image generation and editing by Kling (omni-image, model kling-image-o1). Supports 1K/2K resolution and multi-image input. $0.05 per image.
Grok Imagine Image
xAI
Multimodal AI image generation by X platform. Generates high-quality images from text descriptions.
Grok Imagine Image Pro
xAI
Upgraded multimodal AI model by X platform with stronger understanding and finer detail generation for higher precision images.
SparkPix Image
SparkPix
Sub 1 second text-to-image model built for production use cases. State-of-the-art speed, quality, and text rendering.
SparkPix Image Edit
SparkPix
Sub 1 second multi-image editing model. Fast, affordable AI image editing with precise prompt adherence and multi-image support.
MiniMax Hailuo H3
MiniMax
MiniMax Hailuo H3 — 2K or 768P video with synchronized audio from one multimodal request. Prompt alone for text-to-video, or attach up to 9 reference images, 3 videos and 3 audio clips to steer character, motion and voice. Any length from 5 to 15 seconds, billed per second: 768P $0.097/s, native 2K $0.145/s. Reference video adds $0.135/s (2K) for its own length; reference audio is free.
Grok Video 3
xAI
Grok Video 3. Per-second pricing — $0.04/s at 480p / 720p, $0.12/s at 1080p — with 6 / 10 / 15 second output. Both T2V (omit images) and I2V (1-7 reference images) supported.
Grok Video 3 (10s)
xAI
Grok Video 3 on flat per-clip tiers: 720p $0.30 / $0.50 / $0.70 and 1080p $0.35 / $0.65 / $2.00 for 6 / 10 / 15 seconds. Text-to-video and image-to-video (up to 7 reference images), real 1080p.
Seedance 2.0
ByteDance
ByteDance Seedance 2.0 — full multimodal generation from text, a first / first+last frame, up to 9 reference images, reference video, reference audio and web search, with native synced audio. Real-person images can be passed directly (an upstream rejection triggers an automatic asset-library retry). 480p / 720p / 1080p, 4-15s, per-second pricing from $0.092/s.
Seedance 2.0 Fast
ByteDance
The speed- and cost-optimised tier of Seedance 2.0 — same full multimodal capability (text / first+last frame / up to 9 reference images / reference video / reference audio / web search / native audio), real person via the asset library. 480p / 720p, 4-15s, from $0.071/s.
GPT-5.6 Sol
OpenAI
GPT-5.6 Sol is the frontier model in the GPT-5.6 family — OpenAI's highest-intelligence tier (the gpt-5.6 alias routes to Sol), roughly the unsuffixed top tier of earlier GPT-5 families. Reasoning-token support, text + image input, 1.05M-token context, 128K max output, Feb 2026 knowledge cutoff. Call it on /v1/chat/completions or /v1/responses — one key, well below OpenAI list, reachable from China.
GPT-5.6 Terra
OpenAI
GPT-5.6 Terra balances intelligence and cost — the mini tier of the GPT-5.6 family, priced at half of Sol's output rate and about 40% of its input rate. Higher reasoning, fast, text + image input, text output, 1.05M-token context, 128K max output, Feb 2026 knowledge cutoff, reasoning-token support. Call it on /v1/chat/completions or /v1/responses — one key, well below OpenAI list, reachable from China.
GPT-5.6 Luna
OpenAI
GPT-5.6 Luna is optimized for cost-sensitive, high-volume workloads — the nano tier of the GPT-5.6 family, priced at about a seventh of Sol's input and output rates. High reasoning, fast, text + image input, text output, 1.05M-token context, 128K max output, Feb 2026 knowledge cutoff, reasoning-token support. Served as gpt-5.6-luna-max on /v1/chat/completions or /v1/responses — one key, below OpenAI's $0.20 / $1.20 list (10% on input, 20% on output), pay-as-you-go, reachable from China.
Seedance 2.0 Mini
ByteDance
The most affordable Seedance 2.0 tier — same multimodal inputs (text / first+last frame / reference images, video and audio) and native synced audio at 480p / 720p, 4-15s, from $0.044/s. Best for high-volume drafts and cost-sensitive batch generation.
Grok 4.5
xAI
xAI's newest flagship model — leads the industry in coding, non-hallucination rate, agentic tool calling, and instruction following. Supports both non-reasoning and reasoning modes, 500K context. More affordable than xAI official ($2 / $6 per 1M tokens): 36% off input, ~29% off output.
Seedance 2.0 Cinematic
ByteDance
Film-grade edition of Seedance 2.0 — cinematic lighting, mood and camera motion, and it accepts real-person / realistic human reference images directly. Up to 4 reference images for identity-locked image-to-video — ideal for film-grade portrait and character work. Flat per-clip pricing; generation typically takes 60-180s. Use only with consented subjects.
Suno Music (v4 & v5)
Suno
AI music generation — full songs with vocals + lyrics from a one-line idea (inspiration), your own lyrics (custom), or instrumental only. Also sound effects, continue, cover, and voice personas. Each run returns 2 variants.
Dreamina Seedance 2.0
ByteDance
Seedance 2.0, international line — true native resolution up to 4K (10-bit), no upscaling. Full multimodal: text, first/last frame, up to 9 reference images + video + audio, native synced audio. 480p / 720p / 1080p / 4K, 4-15s, per-second pricing.
Dreamina Seedance 2.0 Fast
ByteDance
The faster, more affordable Seedance 2.0 tier — same full multimodal capability (text / first+last frame / up to 9 reference images / video / audio / native audio) at native 480p or 720p. 4-15s, per-second pricing.
Dreamina Seedance 2.0 Mini
ByteDance
The most cost-effective Seedance 2.0 tier — full multimodal (text / first+last frame / up to 9 reference images / video / audio / native audio) at native 480p or 720p. 4-15s, per-second pricing.
VEO 3.1 Full HD
Google VEO 3.1 (standard tier) at true 1080p — the highest-quality VEO on the platform. Text-to-video, or first / first+last frame from images, 8s clip. For drafts at the same 1080p use VEO 3.1 Fast Full HD ($0.088).
Claude Fable 5
Anthropic
Anthropic's high-end public LLM — about 50% less than official. Works with Claude Code & Cursor via the Anthropic API. Ultra-long context, multimodal, complex reasoning, strict safety.
Claude Sonnet 5
Anthropic
Anthropic's previous Sonnet (the newest is Sonnet 5.5) — 1M-token context (default & max), 128K max output, adaptive thinking, and the same tools & platform features as Sonnet 4.6 (Priority Tier not supported). Works with Claude Code & Cursor via the Anthropic API.
AI Lip-Sync
APIMODELS
Turn a portrait image + an audio clip into a talking-head video. The output length follows the audio; billed per second.
Eleven Flash v2.5
ElevenLabs
Ultra low latency model in 32 languages. Ideal for real-time conversational use cases.
Eleven Turbo v2.5
ElevenLabs
High quality, low latency model in 32 languages. Best for developer use cases where speed matters.
Eleven Multilingual v2
ElevenLabs
Most life-like, emotionally rich mode in 29 languages. Best for voice overs, audiobooks, post-production.
Eleven v3
ElevenLabs
Most expressive model with 70+ languages. Supports audio tags like [laughs], [whispers] for emotional control.
ElevenLabs Dialogue
ElevenLabs
Multi-speaker dialogue generation with natural conversation flow. Perfect for podcasts and audiobooks.
Voice Isolator
ElevenLabs
Extract speech from background noise, music and ambient sounds. Clean audio extraction.
AI Dubbing
ElevenLabs
Translate audio/video while preserving emotion, timing and tone. Automatic lip-sync.
Subtitle Eraser
APIMODELS
Remove hardcoded subtitles, burned-in on-screen text, watermarks and corner logos from any video, rebuilding a clean background where they used to be. High-frame-rate OCR detection plus AIGC inpainting for near-invisible results. Let it find the text automatically, or switch to region mode and box exactly what should go — the way to remove a watermark or station bug. Output resolution matches the input (up to 1080p). Billed $0.015 per second of video.
LTX-2.3 Video
Lightricks
LTX-2.3 unified text-to-video and image-to-video. Send a prompt for T2V, or add one reference image for I2V — fast, with 480p / 720p / 1080p output. Billed per second by resolution.
Omni Flash (Stable)
Omni Flash (Stable) — lower-cost, full-suite Gemini Omni video. Text / image (up to 7 refs) / video-to-video, plus reusable voices and consistent characters. 720p / 1080p / 4k, 6 / 8 / 10s, 16:9 or 9:16, optional seed.
Omni Flash (Gemini)
Gemini Omni Flash — unified video generator for text-to-video, image-to-video (1 or 3 reference images) and video editing. 720p / 1080p / 4k, 6 / 8 / 10s, optional 16:9 or 9:16 framing. Per-second pricing: 720p $0.06/s, 1080p $0.065/s, 4k $0.12/s (video edit is billed on the reference clip length).
DreamActor V2 (Motion Transfer)
ByteDance
ByteDance DreamActor V2 motion transfer. Drive any character image with reference video motion, supporting multi-person, anime and pets.
DeepSeek V4 Flash
DeepSeek
DeepSeek V4 Flash — the lightweight, high-throughput, cost-effective member of the DeepSeek V4 family for general chat and basic text. 1M-token context, tool calling, streaming; OpenAI-compatible.
DeepSeek V4 Pro
DeepSeek
DeepSeek V4 Pro — DeepSeek’s high-performance model with top-tier reasoning and agent capabilities, 1M-token context, and full thinking (reasoning_content) output. Tool calling, streaming; OpenAI-compatible.
Qwen3.7 Max
Alibaba
Alibaba Qwen3.7 Max — the most capable Qwen3.7 model: top-tier reasoning and agent ability, 1M-token context, hybrid thinking. Tool calling, streaming; OpenAI-compatible.
Qwen3.7 Plus
Alibaba
Alibaba Qwen3.7 Plus — the best-value Qwen3.7 model: strong reasoning at a fraction of Max’s price, 1M-token context, hybrid thinking. Tool calling, streaming; OpenAI-compatible.
Kling Lip-Sync Video
Kling
Kling AI lip-sync video generation. Frame-level lip synchronization with audio for real humans, 3D and 2D characters.
Kling Lip-Sync TTS
Kling
Kling text-to-speech synthesis with multi-language support, voice cloning, speed control and emotion styles.
GPT-5.4
OpenAI
Frontier model for complex professional work and agentic coding. Served via the Responses API with adjustable reasoning effort (low–xhigh), web search, and function calling.
GLM-5.2
Zhipu
Zhipu GLM-5.2 — a reasoning model with strong function-calling / tool-use, served via the OpenAI-compatible chat-completions endpoint.
GPT-5.5
OpenAI
OpenAI’s advanced reasoning model for agentic coding, knowledge work, scientific research, and complex multi-step task execution. Served via the Responses API with adjustable reasoning effort (low–xhigh), web search, and function calling.
Kling V3
Kling
Kling V3 video generation. Supports 3-15s flexible duration, text-to-video and image-to-video with optional audio.
P-Video
Pruna AI
Fast video generation in ~10 seconds. Text/image/audio-to-video with draft mode for 4x faster previews. Built-in audio generation, up to 1080p 48FPS.
Minimax Speech 2.8 HD
Minimax
Latest high-fidelity TTS by MiniMax (海螺). Predicts emotion and intonation from context for ultra-natural, expressive, personalized speech. Supports voice clone.
Minimax Speech 2.8 Turbo
Minimax
Latest fast, cost-effective async TTS by MiniMax (海螺). Great quality-to-price for high-volume synthesis. Supports voice clone.
Minimax Speech 02 HD
Minimax
High-fidelity TTS by MiniMax (海螺). Predicts emotion and intonation from context to produce ultra-natural, expressive, personalized speech — built for social, podcasts, audiobooks, news, education and digital humans. Supports voice clone.
Claude Opus 4.8
Anthropic
Anthropic's most capable model yet — built to autonomously carry long, complex work end to end. Ideal for big projects, building agents, and high-stakes scenarios demanding top quality and autonomy.
Claude Opus 4.7
Anthropic
Latest Opus model with 1M context, 128K max output, and adaptive thinking — same tools and platform features as Opus 4.6.
Claude Opus 4.7 (Thinking)
Anthropic
Claude Opus 4.7 with extended thinking explicitly enabled for the most complex reasoning tasks.
Gemini 3.5 Flash
GA release. Our most intelligent Flash model — consistent leadership on agentic execution, coding, and long-horizon tasks at scale.
Gemini 3.1 Flash Lite
Most cost-effective multimodal model with fastest performance for high-frequency lightweight tasks.
Gemini 3.1 Pro Preview
Gemini 3.x Pro preview with enhanced reasoning and multimodal input. For Google's newer flagship, see Gemini 4 Argon.
Kling Custom Voice
Kling
Create custom voice profiles from audio samples. Upload .mp3/.wav/.mp4/.mov (5-30s) or reference a video ID.
Kling Face Recognition
Kling
Identify faces in a video and return a session ID and face IDs for Kling lip-sync video generation.
Claude Opus 4.6
Anthropic
Latest Opus model with ultimate performance and reasoning capabilities.
Claude Opus 4.6 (Thinking)
Anthropic
Claude Opus 4.6 with extended thinking capability for the most complex reasoning tasks.
Claude Sonnet 4.6
Anthropic
Claude Sonnet 4.6, an earlier-generation Sonnet balancing quality and efficiency. The newest Sonnet is Sonnet 5.5.
Claude Sonnet 4.6 (Thinking)
Anthropic
Claude Sonnet 4.6 with extended thinking capability for complex reasoning tasks.
Kling Sound Effects
Kling
Generate sound effects from text descriptions. 3-10 second audio with natural quality.
Kling Video-to-Audio
Kling
Auto-generate sound effects and background music for videos. Supports ASMR mode for immersive content.
Kling TTS
Kling
Text-to-speech with multiple voice options. Adjustable speed and multi-language support.
Minimax Speech 2.6 HD
Minimax
High-definition async TTS by Minimax (海螺). Rich expressiveness with natural prosody. Supports voice clone.
Minimax Speech 02 Turbo
Minimax
Fast and cost-effective async TTS by Minimax (海螺). Supports voice clone and pronunciation dictionaries.
Gemini 3 Flash Preview
Fast and efficient multimodal model. Great for quick responses and simple tasks.
Gemini 3 Pro Preview
Advanced multimodal reasoning model with superior capabilities.
Gemini 3 Pro (Thinking)
Gemini 3 Pro with extended thinking capability for complex reasoning tasks.
Claude Opus 4.5
Anthropic
Latest Opus model with enhanced capabilities and improved reasoning.
Claude Opus 4.5 (Thinking)
Anthropic
Claude Opus 4.5 with extended thinking capability for the most complex reasoning tasks.
Claude Haiku 4.5
Anthropic
Fast and affordable model for lightweight tasks. Best for simple queries and quick responses.
Claude Haiku 4.5 (Thinking)
Anthropic
Claude Haiku 4.5 with extended thinking capability for complex reasoning tasks.
Claude Sonnet 4.5
Anthropic
Claude Sonnet 4.5, the Claude 4.5-generation Sonnet with improved performance and efficiency over earlier Sonnets. The newest Sonnet is Sonnet 5.5.
Claude Sonnet 4.5 (Thinking)
Anthropic
Claude Sonnet 4.5 with extended thinking capability for complex reasoning tasks.
Text Embedding 3 Small
OpenAI
Small embedding model, efficient and cost-effective for most use cases.
Text Embedding 3 Large
OpenAI
Large embedding model for higher accuracy and flexible dimensions.
Z-Image Spicy
Z-Image
Z-Image Spicy — text-to-image at any size from 256 to 1536 px per side (width × height, an OpenAI-style size, or an aspect ratio), about 15 s per image. Seed for repeatable output, optional prompt rewriting. Flat $0.015 per image, one image per request.
Z-Image Spicy Pro
Z-Image
Z-Image Spicy Pro — the higher-resolution tier of Z-Image Spicy: text-to-image at any size from 256 to 2560 px per side (width × height, an OpenAI-style size, or an aspect ratio), 12–21 s per image in our tests. Seed for repeatable output, prompt rewriting off by default. Flat $0.03 per image, one image per request.
Seedream 5.0 Pro Spicy
Doubao
Seedream 5.0 Pro Spicy — the relaxed-moderation variant of Seedream 5.0 Pro, with the same photorealistic quality and the same request shape. Text-to-image and image-to-image with up to 10 reference images, 1K / 2K output (no 4K), JPEG. $0.05 per image at 1K and $0.10 at 2K, charged only on success; typically 40–80 s per image.
Qwen Image Edit Spicy
Alibaba
Qwen Image Edit Spicy — edit an image with a text instruction. Send the image to edit plus up to 2 optional reference images (images 2 and 3 in your prompt); the result keeps the exact pixel size of the image you send. Typically 8–20 s per image. $0.05 with the main image only, $0.06 with 1 reference, $0.07 with 2, charged only on success.
