All models. One API.

The top tier of OpenAI's GPT-5.6 generation. Frontier reasoning, coding, and agentic work with a 1M-token context and image input.

The balanced tier of OpenAI's GPT-5.6 generation. Same 1M-token context and image input, tuned for everyday production work.

The fast, high-volume tier of OpenAI's GPT-5.6 generation. Keeps the 1M-token context and image input at the lowest rate in the family.

Moonshot's new flagship generation. A 1M-token context and the strongest measured reasoning in our catalog.

MiniMax's top video tier. 5 to 15 second clips at 480P or 768P with native audio, driven by a prompt, first and last frames, or up to 9 reference images.

MiniMax's newest video model with native audio. 4 to 15 second clips at 768P or 2K, priced per second.

The official DeepSeek V4 Flash build. Same size and price as the preview, substantially stronger on agentic and coding work.

OpenAI's flagship GPT-5.5. Top-tier general intelligence, coding, and agentic tasks with a 1M-token context.

Google's most capable Flash yet, built on 3.7 for software engineering and autonomous agents. Tops Terminal-bench 2.1 at 89.4% and lifts DeepSWE to 73.7%, with a 1M-token context at the same price as 3.7.

Google's newest fast Gemini, built for coding and agents. Sharp gains over 3.6 on real engineering tasks, the fastest output speed of any model measured, and a 1M-token context.

Google's fast Gemini 3.5 Flash. Near-frontier quality for agents and coding at high speed and low cost, with very long context.

xAI's flagship Grok 4.3. Fast, low-hallucination reasoning with strong agentic tool-calling, 1M context, and real-time knowledge.

Moonshot's Kimi K2.6. Trillion-parameter MoE for strong agentic coding and tool-use at a low price, with 256K context.

Anthropic's newest frontier model. Highest capability for complex reasoning, long-form, and agentic tasks.

Google's frontier Gemini model. 1M-token context, strong reasoning, long-document analysis, coding, and multilingual support.

DeepSeek's next-generation flagship. 1M context, native reasoning mode, strong agentic coding and STEM.

DeepSeek V4's fast variant. Same 1M context and reasoning mode, optimised for high concurrency and low latency.

Zhipu's Flash tier on the GLM-5.3 base. A 1M-token context, native image and video input, and reasoning on by default.

Nano Banana Pro: Google's Gemini 3 Pro Image, frontier text-to-image with best-in-class in-image text.

The top tier of the Seedream 5 line. Single-image output up to 2K, with multi-image fusion for style and character references.

OpenAI's image model, billed by the tokens it actually generates. Strong prompt-following and legible text in images.

ByteDance's Seedance 2.5: 30-second single shots, up to 50 reference assets, in-video editing and extension, native Arabic prompting and speech.

The faster, ~20% cheaper tier of Seedance 2.0: same multimodal video generation, capped at 720p.

ByteDance's Seedance 2.0: multimodal video generation (text/image/reference → video) with optional synchronized audio.

Meta's compact 8B Instruct: strong baseline for high-throughput backend chat and agent loops at near-zero cost.

Alibaba's dedicated machine-translation model. Tuned for accuracy on Arabic↔English and 90+ language pairs.

Flagship vision-language model from Alibaba. 235B MoE with active-22B routing, 131K context, strong image understanding.

Mature 72B dense vision-language model. Baseline VLM tier: different price-quality point from the 235B flagship.

Industry-standard multilingual embedding. Dense, sparse, and multi-vector retrieval in one model. The RAG default.

Tiny 0.6B Qwen3 embedding: designed for high-throughput RAG with a 32K window. Arabic-strong alternative to BGE.

Industry-standard multilingual reranker. Drop in as the second stage of any RAG pipeline behind a vector recall step.

Prompt-driven image editor from Alibaba. Single-pass edits like 'add sunglasses', 'change background to beach', 'remove the person on the left'.

Mid-tier FLUX.1 Kontext: image edit and text-guided generation. Slots between Schnell (fast/cheap) and Kontext Max (premium).

Kling 3.0 in 4K: text-to-video at premium resolution. Different SKU from the standard-resolution V3.0 Pro.

Kling 3.0 in 4K: image-to-video. Animate a still in premium resolution for hero / campaign output.

Current-generation MiniMax Hailuo: text-to-video. Sharper motion and longer coherence than the Hailuo 02 predecessor.

Image-to-video variant of Hailuo 2.3. MiniMax's first I2V model: animate stills with prompt-controlled motion.

Latest-gen Wan: text-to-video. Per-second pricing at $0.10/s makes it the cost anchor for long-form video.

Latest-gen Wan: image-to-video. Same $0.10/s pricing as T2V: the budget default for animating stills.

Clone any voice from a 30-second reference sample. The premium tier for voice replication: MiniMax leads the quality bar in this category.

Design a custom voice from a text description: 'middle-aged Arabic male, warm tone, slight raspy edge'. Companion to Voice Cloning.

Cheap voice cloning at $0.10/voice: 24× cheaper than MiniMax. Trade quality for cost; use for higher-volume cases.

Text-to-music generation. Describe the song you want: genre, mood, instrumentation, lyrics, and the model composes and renders a track.

Google's newest fast Gemini. Near-frontier quality for agents and coding at high speed, now more token-efficient with a lower output price than 3.5, plus a very long 1M-token context.

Frontier-tier general intelligence with strong reasoning, coding, and multilingual coverage.

Anthropic's workhorse coding and chat model. Best Arabic among Western flagships, strong tool use, 200K context.

Alibaba's flagship MoE: 397B total / 17B active. Frontier general intelligence with the best Arabic in the catalog.

OpenAI's flagship. Frontier reasoning, vision input, 400K context. Tier 1 pricing up to 272K.

Anthropic's frontier model. Highest capability for complex reasoning, long-form, and agentic tasks.

xAI's Grok 4. Real-time-aware reasoning model with strong coding and math.

Meta's Llama 3.3 70B Instruct. Open-weight foundation, widely fine-tuned, reliable baseline.

Open chain-of-thought reasoning. Competitive on math, coding, and logic benchmarks against frontier reasoning models.

Kimi K2.5: Moonshot AI flagship. 200K context, strong agentic and tool-use performance.

V3.1 with the Terminus refresh: improved coding, longer effective context, lower hallucination rate.

Google text generation model. Available via THALAM.

Zhipu's flagship vision-language model. Strong document, OCR, and chart understanding.

GLM-5: the prior generation. Cheaper than 5.1, still very capable for general use.

Zhipu's flagship GLM 5.1: long context, strong multilingual coverage, reasonable pricing.

GPT-5 mini: smaller, faster sibling. Capable and well-suited to high-volume production tasks.

OpenAI's open-weight 120B coding model. Apache-licensed, fully self-hostable.

MoonshotAI text generation model. Available via THALAM.

Llama 4 Maverick: Meta MoE flagship. Open-weight, 128 experts, FP8 quantized for speed.

Llama 4 Scout: efficient sibling. 16 experts, lower cost, still very capable.

M2.5: the slightly older MiniMax. Same pricing as M2.7, kept for backward-compat.

MiniMax M2.7: fast multilingual chat. Strong Asian languages, decent English, cheap.

Cheapest model in the catalog. Mistral-Nemo for budget pipelines.

Qwen3 Max: long context, strong reasoning, tiered pricing for high-volume use.

Mid-tier Qwen3.5 MoE. The price/performance sweet spot for production workloads.

IDE-tier coding model. Cheap, fast, fits inline-completion workloads.

Qwen3 Coder 480B: Alibaba's flagship coding MoE. 256K context, 100+ languages, fill-in-the-middle.

Alibaba coding model. Available via THALAM.

FLUX 2 Pro: frontier text-to-image. State-of-the-art aesthetics, prompt fidelity, fine detail.

FLUX.1 Kontext Max: image edit and conditional generation with text prompts.

Seedream 4.0: best image deal in the catalog.

Seedream 4.5: refresh of 4.0, same anchor pricing, slightly better fidelity.

Seedream 5.0 Lite: newest generation, smaller checkpoint, fast.

Z Image Turbo: cheapest image model, $0.005/image price anchor.

Hunyuan Video Fast: Tencent’s speed-optimised T2V at $0.30/video.

Kling V3.0 Pro: image-to-video. Animate stills with the same engine as T2V.

Kling V3.0 Pro: Kuaishou's flagship text-to-video. State-of-the-art motion fidelity.

Kuaishou video generation model.

MiniMax Hailuo 02: fast, affordable text-to-video.

ByteDance video generation model.

ByteDance video generation model.

Wan 2.5 T2V Preview: best video deal in the catalog.

Wan 2.6 T2V: Alibaba’s text-to-video. Strong on cinematic shots and long takes.

ElevenLabs v3: gold-standard voice synthesis.

ElevenLabs Multilingual v2: faithful, consistent voice synthesis.

Fish Audio TTS: multilingual voice synthesis with cloning.

MiniMax audio model (TTS / STT).

MiniMax 2.8 Turbo: latency-optimized variant for real-time TTS use cases.

MiniMax Speech 2.8 HD: top-tier multilingual TTS with natural prosody.