Checked against the OpenRouter public models feed. Prices are copied, not estimated.

The Best Model ForWhich AI model to use, from ten labs.

Ten labs

Short answer

The same ten companies, one page each. Each page leads with three reviewed models, a real task for each, and the full current list underneath with prices from the feed.

OpenAI

OpenAI sells the GPT family as a hosted API. On today's feed the current general models are GPT-6.2 Astra, GPT-6.2 Sol, and the cheaper GPT-6 Luna, plus older GPT-5 snapshots, audio models, and image models. Astra Pro is listed as the same model as Astra with a higher reasoning mode, not as a separate brain.

79 rows on the feed.

Anthropic

Anthropic's Claude models on this feed are hosted APIs with text, image, and file input. The practical ladder is Fable for the hardest technical work, Sonnet for everyday coding and writing, and Haiku for high-volume drafts. Opus sits between Fable and Sonnet on price.

16 rows on the feed.

Google (Gemini)

Google's listing mixes Gemini 4 chat models, older Gemini 3 Flash models, open-weight Gemma models, and separate image, speech, and video models. Gemini 4 models accept text, images, audio, video, and files. Gemma rows carry a public weights id.

49 rows on the feed.

xAI (Grok)

xAI's Grok models are listed under the SpaceXAI name on OpenRouter. Grok 4.7 is the current flagship. Grok 4.20 publishes the longest context in this set. Grok Build is the coding specialist. Image and video generators are separate rows and are not priced like the chat models.

12 rows on the feed.

Meta (Llama)

Meta's current hosted lineup on this feed is the Muse family. The open-weight Llama models are still listed, including Llama 4 Scout and Maverick. Muse Wonder is the frontier hosted model. Llama 4 Scout is the cheap, long-context, open-weight option, with a knowledge cutoff the feed dates to 2024.

18 rows on the feed.

Mistral

Mistral publishes many of its models with a Hugging Face weights id, so open weights and a hosted price often appear on the same row. Large 4 is the frontier multimodal model. Devstral 2 is the coding specialist. Small 4 is the everyday cheap model. Voxtral rows cover audio.

30 rows on the feed.

DeepSeek

DeepSeek's rows on this feed include public weights ids. The October 2026 V4.1 Flash snapshot is the reviewed default: it accepts images and is listed cheaper than the April V4 Pro row. The name Pro does not, on this feed, mean a higher third-party index.

14 rows on the feed.

Alibaba (Qwen)

Alibaba's Qwen listing runs from the hosted Qwen4 Max flagship to open-weight Flash and coder models. Several rows accept video as well as images. Qwen3 Coder Next is the open-weight coding agent. Image, speech, and embedding models are listed separately.

52 rows on the feed.

Cohere

Cohere's feed is smaller: Command A+ for enterprise tool use, the older open-weight Command A, and the small Command R7B for retrieval and tools. Command A is listed at a higher token price than Command A+, so open weights are not a discount here. Embed and rerank models are listed too.

12 rows on the feed.

Amazon (Nova)

Amazon's generative rows here are the Nova family. Premier is the expensive reasoning model. Nova 2 Lite is the everyday model and is the one that lists video input. Nova Micro is text-only and the cheapest reviewed Nova. Titan rows are embedding models, not chat.

8 rows on the feed.