AI Stats

AI Stats is an open-source, all-in-one AI platform for enthusiasts and developers. Follow AI data, trends and insights, or build with 300+ models through our AI Gateway, with more added weekly. Built to democratise AI through openness, accountability and community. Check it out here: ai-stats.phaseo.app

@MistralAI Pricing starts at $1.5 Per Million Input Tokens and $7.5 Per Million Output Tokens. Give it a go right now and let us know how you find it: ai-stats.phaseo.app/models/mistral…

@MistralAI Medium 3.5 is now live on AI Stats! Try it in our chat, test it on our unified API alongside over 300 other models, and see the latest benchmark results compared to other frontier models! Our whole platform is open source, so if you have any suggestions, please let us know!

This is a good question! From what I have found - DeepSeek have put a lot of work into making their models super efficient and cache is crucial to that, obviously different use cases cache differently. For example, using it inside of a Code Harness like OpenCode will lead to high cache hit rates due to files not changing frequently, however, other use cases can cause the cache to be invalidated more frequently, as AI is heavily used in programming currently, the cache rates generally remain quite high, however, it does largely depend on the use cases as to how cache efficient the models will be. With the discount right now though from DeepSeek, it is too good of a deal not to give the model a go and see how it works in your use case!

@ai_stats_team It is claimed that DeepSeek V4 has a very high cache hit rate, I wonder how it is achieved?

DeepSeek cache pricing for DeepSeek V4 Flash and V4 Pro has been updated! 90% reduced cache hit pricing, stacked with the 75% discount on DeepSeek V4 pro too! Give it a go right now through our AI Gateway!

Even at the weekend, DeepSeek giving us presents: DeepSeek V4 Pro is now on a 75%, yes you read that right, 75% discount until May 5th! This discount is now live on @ai_stats_team! Try out DeepSeek's best model at 75% off! Let us know what you are gonna build!

GPT 5.5 and 5.5 Pro now available in the OpenAI API, coming online on @ai_stats_team within the next 15 minutes, excited for you to try out OpenAI's latest models, and test them in your workflows!

DeepSeek V4 is here! Live on AI Stats RIGHT NOW! Try it out with multiple providers, with more coming online throughout today! You can even take advantage of @gmi_cloud launch discount of 20% through us! Give it a go and let us know how you find it!

DeepSeek V4 Flash/Pro are now live! Don't rely on just one provider whilst testing these models, access both across leading providers @atlas_cloud_ai and @gmi_cloud as well as official @deepseek_ai providers, with more coming online as they release today! Excited to see you build with DeepSeek's new frontier model! Please let us know if you find any bugs!

GPT Image 2 from @OpenAI is officially here and live on AI Stats, go and create with the worlds best Image Model right now on AI Stats! Get Started Here: ai-stats.phaseo.app/models/openai/…

Claude Opus 4.7 has been fully rolled out, get started today with @AnthropicAI's latest model! Scoring over 10% higher in SWE Bench Pro compared to Opus 4.6 it should be even better at your hardest coding tasks! Get started here: ai-stats.phaseo.app/models/anthrop…

Opus showing a marked improvement here, excited to hear your feedback when using it!

@claudeai Give it a go on ai-stats.phaseo.app and let us know how you find it! Rolling out now!

Introducing Claude Opus 4.7, our most capable Opus model yet. It handles long-running tasks with more rigor, follows instructions more precisely, and verifies its own outputs before reporting back. You can hand off your hardest work with less supervision.

Qwen release Qwen3.6 35B A3B! Showing impressiver performance even compared to Gemma 4! Very impressive, can't wait to play around with this and get it onto AI Stats Gateway!

Muse Spark will arrive on AI Stats as soon as it releases! Looking forward to getting this model available!

Today we're releasing Trinity-Large-Thinking. Available now on the Arcee API, with open weights on Hugging Face under Apache 2.0. We built it for developers and enterprises that want models they can inspect, post-train, host, distill, and own.

@rohanvarma Yes for sure! For overnight tasks, I would definitely use it!

If we made /slow mode in Codex, would you use it? What for? (Slower inference at a cheaper cost)

Looks like Veo 3.1 Lite is coming today!

Novita becomes our first major provider to add to the API - we hope to add many more over the coming weeks, and look forward to announcing more options, fallbacks and new capabilities alongside these providers

AI Stats x Novita. Today we are announcing Novita as a supported provider on AI Stats! Novita has around 70 models now accessible through the AI Stats Gateway, most notably MiniMax M2.7 and GLM 5 as well as the GPT OSS models and many more!