Use your own AI models on the web
Joined April 2023
- Tweets113
- Following84
- Followers487
- Likes196
Pinned Tweet
A project for decoupling web apps from AI models. Users choose the model instead.
Demos and download instructions here:
Window AI retweeted
The best evals are your own. Compare Llama, Sonnet, and GPT-4o with a simple group chat here:
According to the @scale_AI SEAL eval: Sonnet > Llama 405B > GPT-4o
Group chat with all of them here! openrouter.ai/chat
Stay tuned for some updates coming 🔜 😏
Our years-old dream of seeing awesome LLM-powered games is slowly coming true!
Didn't happen on top of @windowai_io, but it instead on @openrouter
Try out WebSim if you haven't yet - it's the start of something special.
Window AI retweeted
Two records last week for @openrouter:
- Record # of models added in one week (12)
- Record # of tokens processed (52 billion)
Launching 3 new models, including a Llama 3 finetune (free for a limited time).
Focusing on performance and dynamic routing this week.
New Models 👇
- Soliloquy 8B, a Llama 3 roleplay finetune
- Fimbulvetr 11B, a highly-acclaimed roleplay model based on Solar
- Llama 3 8B Extended to 16k context. Might not last forever, depending on usage
Window AI retweeted
Ever wondered which apps are using an LLM? Now you can find out yourself with the new Apps tab.
@NousResearch has the top Claude 3 Opus apps this week 👀
Window AI retweeted
Just crossed 100 million completions 🪅
See live statistics about 85 LLMs, along with latency and throughput for different hosts. More on the way: openrouter.ai/rankings
Window AI retweeted
Wow! Just launched it in our playground, with a free API: nitter.cf/OpenRouterAI/status/17…
One of the top Transformer alternatives is maturing, and now competitive with Mistral!
Use it for free on OpenRouter until April: openrouter.ai/models/recursa…
Window AI retweeted
2/4: 🧡 Mistral Medium, Small, and Tiny (all 32k context)
We've added support for @MistralAI's API-based models! Mistral Medium is showing some promising results, comparable with flagship models from others.
Now you can get:
- USD prices, without currency conversion
- Compatibility with OpenRouter credits
- Credit-based rate limiting
The Small model compares with Mixtral 8x7b. Check out the Medium model here: openrouter.ai/models/mistral…
Window AI retweeted
Lots of updates: new Mistral models, big price cuts, and more!
1/4: 🤑 Price Cuts
- MythoMax 13B: 63% price cut, down to $0.00022/1k tokens
- Remm Slerp: 80% price cut, down to $0.00022/1k tokens
That's 4.55M tokens per dollar, and less than 1/4th the price of GPT3.5. Check out stats here: openrouter.ai/models/gryphe/…
Window AI retweeted
Just launched an update to our Mixtral 8x7b Instruct router (still free!), which makes it 11x faster than before with 8x the context length.
We also made our avg routing time 75% faster by improving edge tokenization. More perf improvements coming soon! openrouter.ai/models/mistral…
(The previous Primary had an attention issue, so it's been reduced to 4k context, h/t @alexsablay and @WolframRvnwlf)
Window AI retweeted
Our principles:
1. LLM prices are racing to the bottom. Make sure you benefit.
2. Use a Standardized API across models & providers
3. The best models will be used the most. Evals are flawed. Power users will figure out the best models for each use case.
openrouter.ai/docs
Window AI retweeted
Mixtral is one of the fastest growing models we've seen.
Play with it for free here: openrouter.ai/models/mistral…
Window AI retweeted
Not only are AI models commodities — they’re also increasingly hosted by multiple providers, particularly open source models.
This means that a market is emerging between sellers (model hosts) and buyers.
What’s the best way to get market efficiency on this? In my view, a centralized orderbook really helps, like @openrouter. The sellers will eventually decentralize though, and zkml or optimistic validation (like @ritualnet or @bittensororg ) could be ways of doing that.
Non consensus view: AI is already a commodity. OpenAI was not as good as Google’s best internal stuff (I know, was there, worked with teams building Google’s stuff). But they shipped faster and better.
No one is 10x better, and OSS models are following fast. The winner will win by traditional techniques - better UI, better distribution, better marketing, etc.
Window AI retweeted
The OpenAI saga created a big week for open models...
8 billion tokens processed on OpenRouter across 2000+ apps & 52 chat models 🛫
ALT https://openrouter.ai/rankings
Window AI retweeted
Launching three new models tonight:
MythoMist 7B, from the maker of MythoMax: openrouter.ai/models/gryphe/…
Noromaid 20B: a collab by @Undi95 and IkariDev, getting rave reviews and strong when paired with specific instruct settings in ST: openrouter.ai/models/neversl…
Intel's Neural Chat 7B: a new big corp is entering the arena with a top-ranking 7b model: openrouter.ai/models/intel/n…
Window AI retweeted
Excited to open-source our LLM inference engine!
Based on @modal_labs and vLLM, it supports YaRN and RoPE context extension, and provides fallback hosting for many of the 50+ models on OpenRouter.
👉 github.com/OpenRouterTeam/op…
Window AI retweeted
All-in on frankenmerges 🧌 Should we make a bounty for new ones?
Ping us in Discord if you're building one that works, and we'll feature it!
I've become a big fan of "frankenmerge" LLMs that combine weights of open-source models and (inexplicably) result in something better than the sum of the parts. Feels like a very underexplored area of ML.
Examples: Toppy by @Undi95 is now top of the fastest growing models on @openrouter, above GPT-4 Turbo, and MythoMax (red bar) has outpaced everyone. Goliath 120b is also new this weekend.
Here's a search to track them all: openrouter.ai/models?q=%23me…
Window AI retweeted
.@teknium's OpenHermes 2 is now the fastest-growing model on OpenRouter, by tokens processed:
Window AI retweeted
Announcing our first roleplay data bounty 🧙
Submit your favorite AI roleplay chat sessions and get a change to win credits usable with any model on OpenRouter (including OpenAI, Claude, etc):
🥇 First place: $300
🥈 Second place: $200
🥉 Third place: $100
4th and 5th place both get $50.
We'll judge them based on totally arbitrary factors: how emotionally gripping / insightful it is, how creative the user is, and how creative the AI is. Group chat submissions encouraged.
Rules:
1. Multiple submissions per person are allowed.
2. You must include the app you used (e.g. Character AI, SillyTavern, etc) and the model, if you used a bring-your-own-model UI like ST
3. If we believe a submission was doctored and not actually created by AI, it's disqualified.
4. Deadline to submit: Friday Nov 10. If we don't receive at least 10 good submissions by Thursday, it may be extended by one week.
5. To submit, send an email from your OpenRouter account's email address to "support {-at-} openrouter.ai"