Modelle

The strongest, safest OPTES: planning, coding and review in one model. Up to 1,048,576 tokens of context.

Fast and low-cost for chat, classification, summaries and high-volume requests that cannot wait.

A coding agent that reads the whole repository, writes, runs the tests and reviews what it changed before handing over.

Understands images and documents, reads printed and handwritten Arabic, and extracts tables and fields.

Images from a written description in Arabic or any language, in three quality levels (from $0.006 fast to $0.04 HD).

Video clips with their sound from text or an image (from $0.08 a second, 720p without audio, to $0.35 at 1080p with audio).

Natural speech in Arabic, its dialects and other languages; your own cloned voice only with documented consent.

Songs with your lyrics, or instrumentals for a video or a game, at the length and mood you choose.

Speech to text that understands Arabic dialects, separates speakers and timestamps every sentence.

Turns text into vectors so you can search your documents by meaning, not by word.

Re-orders search results by their real relevance to the question, so the most important comes first.

Checks inputs and outputs before they arrive and flags what breaks your policy or ours. Free with any use.

Anthropic's model for demanding reasoning and long-horizon agentic work.

OpenAI's flagship on its Models page; requests above the short-context tier cost more.

Anthropic's recommended default for most work: long-running agentic coding and knowledge work.

The newest Flash. Promotional price through 31 December 2026; the price doubles from 1 January 2027.

Z.ai's flagship GLM, hosted by more than a dozen providers.

The mid-size GPT-6.1 model: most of the flagship's ability at a fifth of its price.

xAI's strongest model per its Models page; prompts over 200K tokens cost double at the provider.

Moonshot's flagship with a 1M-token context; one of the most widely hosted models in the catalog.

The best combination of speed and intelligence in the Claude line.

DeepSeek's current flagship (the 0813 release): 1M context and up to 384K output; half price off-peak on DeepSeek's own platform.

The fast multimodal GLM-5.3, released under MIT.

The smallest GPT-6 model, for high volume at a very low price.

MiniMax's long-context agent model. In OPTES plans it reads up to 524,288 tokens; MiniMax's own platform offers 1M, at double the price above 512K.

Alibaba's largest Qwen (2.4T total, 95B active).

DeepSeek's fast model (deepseek-flash on its own API): 1M context and image input.

Google's most capable Gemini, in preview; priced for prompts up to 200K tokens.

The earlier GPT-6 Sol, listed on Azure AI Foundry.

Grok 4.6 as listed on Azure AI Foundry.

The fast, low-cost Qwen3.8.

xAI's model for building software.

xAI's image generator.

Moonshot's coding model with a 262K-token context.

The 27B open member of Qwen3.8.

Gemini 3.5 Flash, the earlier Flash generation.

The previous GLM generation.

A lower-cost Grok with a 1M-token context.

The faster K2.7 Code, at twice the price.

The newest Muse Spark on Meta's Model API (public preview), with a 1M-token context.

StepFun's multimodal reasoning model, in preview.

The 0731 release of V4 Flash.

The lightest Gemini 3.5 model for volume work.

The K2.6 generation, also offered as K2.6 Thinking.

A much cheaper Muse Spark 1.3 variant on Meta's Model API.

GLM-5.1.

Meta's image generator.

The previous Opus generation, still available.

The original V4 Pro release, still hosted by many providers.

Tencent's newest Hy model, in preview.

Xiaomi's MiMo Pro at a low international price.

Meta's speech-to-text model.

Qwen3.7 Plus, priced without thinking up to 256K tokens.

StepFun's fast Step 3.7.

The original V4 Flash: one of the cheapest capable models in the catalog.

GLM-5.

Xiaomi's fast MiMo.

The high-speed M2.7.

A GPT-5.6 model; the only price in the research is an older aggregator figure.

MiniMax's text-to-video model.

MiniMax M2.7 with a 204.8K-token context.

Mistral's newest Medium, released with open weights (modified MIT).

The 27B open member of Qwen3.6.

A small Qwen3.6 mixture-of-experts (3B active).

ByteDance's Seed 2.0 Pro, priced for prompts up to 128K.

StepFun's cheapest fast model.

Mistral's small, low-cost model.

The small Seed 2.0.

GPT-5.5, priced for prompts under 272K tokens.

Tencent's Hy3 at a very low price.

Mistral's large open-weights model, offered by Mistral and the big clouds.

GPT-5.4 as listed on Azure AI Foundry.

The 8B member of Ministral 3, at the same price for input and output.

NVIDIA's small fast Nemotron (30B total, 3B active).

Mistral's model for code.

The earlier K2.5 generation.

MiniMax M2.5.

NVIDIA's largest open Nemotron (550B total, 55B active).

Writer's newest Palmyra with a 1M-token context.

The large Qwen3.5 mixture-of-experts (17B active).

Upstage's newest Solar; promotional price through 11 October 2026.

A small 9B Qwen3.5.

The small Solar 4; promotional price through 23 October 2026.

Amazon's Nova 2.0 Pro on Bedrock.

The fastest Claude model with near-frontier intelligence.

Z.ai's low-cost GLM-4.7 FlashX.

Kling's professional video model.

The earlier Medium 3, on Vertex AI.

The previous Palmyra (1M context); deprecated by Writer on 14 December 2026.

Arabic-first model from Inception (G42) on Core42 Compass.

Perplexity's search-grounded answer model.

GLM-4.7 Flash (free on Z.ai itself, paid at hosts).

An earlier Sonnet, offered on Core42 Compass.

MBZUAI's reasoning model from Abu Dhabi.

Reka's smallest, cheapest model.

Google's open-weights Gemma 4 at 31B parameters.

Reka's mid model that reads text, images, video and audio.

AI21's small Jamba model.

The previous DeepSeek generation.

OpenAI's open-weights model, served by more providers than almost any other.

IBM's open Granite 4 hybrid model.

AI21's large Jamba model.

Meta's open-weights Llama 4 mixture-of-experts model (17B active).

Reka's largest model; images and video are billed per item at the provider.

inference.net's model for turning text into structured data.

The small open-weights gpt-oss model, light and cheap to run.

The smaller Llama 4 (17B active, 16 experts).

DeepSeek V3.1.

Qwen3 Next at 80B.

Qwen's large coding model (35B active).

Amazon's smallest, cheapest Nova.

Cohere's multilingual line (8B and 32B), one price for both sizes.

Deep Cogito's 671B open model.

The previous-generation Pro, priced for prompts up to 200K tokens.

GLM-4.5.

The July 2025 Qwen3 235B instruct model.

The July 2025 Qwen3 235B thinking model.

Upstage's embedding model.

Previous-generation Flash, offered on Oracle Cloud.

The lighter GLM-4.5 Air.

Previous-generation Flash-Lite; reads text, images and video at one input price.

A small Qwen3 mixture-of-experts (3B active).

The dense 32B Qwen3.

DeepSeek's first reasoning model.

Meta's widely hosted open 70B model.

Mistral's embedding model.

R1 reasoning distilled into a 14B Qwen model.

Fast open image model from Black Forest Labs, billed per megapixel.

The older small GPT-4o model, still offered through routers.

Meta's small open 8B model.

The faster, cheaper Whisper large-v3.

Open speech-recognition model for transcription in many languages.