OpenRouter

Other launches from us this week: new provider routing settings, APIs, and 10 new models, including: - a free Llama 3.1 8b - Llama 405B base, making a big splash this weekend - New online models from Perplexity - Yi models, including Yi Large, Yi Vision, Yi function calling Want to avoid having any your requests routed to certain providers? Now you can configure this on your settings page: https://lnkd.in/eWKq4H2T You can now use our Parameters API to check for the parameters and settings supported for a given model (and provider!) https://lnkd.in/eDrFnTGJ Llama 3.1 405B BASE! It got its own tweet here: https://lnkd.in/eu5JzGdc People are using it to resurrect Sydney: 🍕 New Freebie: Llama 3.1 8B Use the new stable small open source model for free here, both in the Chatroom and via API: https://lnkd.in/eSjkB7UB Mistral Nemo 12B Celeste A specialized story writing and roleplaying model from nothingiisreal, based on Mistral's NeMo 12B Instruct. https://lnkd.in/excGAFk2 Llama 3.1 Sonar family from @perplexity_ai Try these new 🌐 online models for helpful, up-to-date, and factual responses: Llama 3.1 Sonar 70B Online: https://lnkd.in/ezzEvzpp Llama 3.1 Sonar 8B Online: https://lnkd.in/eUdmkKi9

📉 14% Price Cut for Llama 3.1 8B The price wars continue on, just a few days after launch. Where will it settle? https://lnkd.in/gz635aYw

DeepSeek Coder V2 now has a private provider serving requests on OpenRouter, with no input training! Check it out here: https://lnkd.in/eCi7Z4PN

Llama 3.1 405B has arrived! Rivaling GPT-4o and Claude 3.5 Sonnet, it's live now: - $3/M tokens (will likely fall over time) - 128K token context - Allows synthetic data generation for training other models! As always, more providers will be added as fast as we find them. 👇 Llama 3.1 405B APIs are here: https://lnkd.in/evH6YVSf Llama 3.1 8B and 70B are on the way

New models from @MistralAI have arrived. 1. Mistral Nemo: A 12B parameter multilingual LLM with a 128k context 2. Codestral Mamba: A 7.3B parameter Mamba-based model designed for code and reasoning Try them out here! https://lnkd.in/e95BAand https://lnkd.in/eGCGWe-K

Single LLM providers aren't always online! During an outage one had today, our router recovered and fulfilled 30k requests that would've otherwise failed. See similar charts by visiting a model here and clicking "Uptime": openrouter.ai/models

🎁 New Models! Magnum 72B by @AlpinDale: https://lnkd.in/eVFQEYnY From the maker of Goliath, Magnum is the first in a new family of models designed to achieve the prose quality of the Claude 3 models, notably Opus & Sonnet. The model is based on Qwen2 72B and trained with 55 million tokens of highly curated roleplay (RP) data. Hermes 2 Theta by @NousResearch: https://lnkd.in/eQnt2W_s An experimental merge model based on Llama 3, exhibiting a very distinctive style of writing. It combines the the best of Meta's Llama 3 8B and Nous Research's Hermes 2 Pro. Hermes-2 Theta was specifically designed with a few capabilities in mind: executing function calls, generating JSON output, and most remarkably, demonstrating metacognitive abilities.

Announcing a brand-new UI for discovering language models✨ Explore 180+ active language models processing 74 billion tokens/week on the first and largest LLM marketplace 👇 - Looking for just multi-modal models? now you can compare them directly, seeing open and closed source models together: https://lnkd.in/eM7d63Ug - Just want the freebies? here they are (15 and counting): https://lnkd.in/e9nRnr3v - Want to see how different model series compare? See all Llama 3 vs Llama 2 vs Yi vs Mistral models in one place: https://lnkd.in/eDZ96NM8 - Just want to see models good at programming or roleplay? Try the category filter: https://lnkd.in/eBKQj7ac - Last but definitely not least: you want models with tool calling or JSON support or min_p? Now you can filter for 16 features in one spot. Example: tool calling models: https://lnkd.in/eAYCYUMR

Top trending programming model this week: StarCoder2 by @BigCodeProject It's their first self-aligned open-source LLM, fine-tuned without any human annotations or distilled data from proprietary LLMs.

The new Qwen 2 72B is now available, with top-tier performance among open source models. Already 44M tokens, and 95+ tokens per second throughput. Try it here: https://lnkd.in/euB8rA2U