Log in Subscribe
Newsletter

🚨 One of the biggest shake-ups in AI leaderboards

Alexey Shabanov
· 10 min read

Greetings!

This week was one of the biggest we have seen in terms of AI releases, especially new model launches. Alongside the expected release from OpenAI, we also saw a somewhat anticipated launch from xAI and a surprising release from Meta Superintelligence Labs.

Beyond the new models, ChatGPT and Codex were finally merged. However, there were also some delays, with Seedance 2.5 reportedly pushed back by around one week.

What is particularly interesting is that the industry has clearly shifted from quarterly release cycles to monthly ones. The next question is whether we will reach biweekly releases by the end of the year. Release frequency is a useful indicator for anyone tracking the pace of AI development and wondering when a potential exponential takeoff might become visible publicly. Once AI labs learn how to upgrade their models more frequently, and begin treating that as a core objective, it could lead to substantial acceleration in progress.

Whether this is realistic and whether it will actually happen remains to be seen. But for now, let’s dive in!

Newsletter Guide

Here's your guide on how to make the most out of this newsletter:

  1. 🤳 Gear up with your mobile phone and desktop devices.
  2. 📩 Scan through the newsletter, picking out the apps that pique your interest.
  3. 🔥 Keep an eye on the "hot" emoji to identify the hottest user experiences.
  4. 👀 Dive deeper into our TestingCatalog posts to discover how to put these apps to the test.
  5. 📲 Roll up your sleeves and try them out yourself!

This weekly newsletter was sponsored by:

1X

Anthropic

On the Anthropic side, things were a little quieter this week. The company released Cowork for web and mobile, which was well timed given that OpenAI moved in a very similar direction this week. Users can now manage Cowork tasks while on the go.

Anthropic also extended the inclusion of Fable 5 in its subscription until July 12, which is today. However, it appeared to time the reset of weekly usage limits around the launch of GPT-5.6, likely to capture some of the attention surrounding OpenAI’s release. That is not particularly surprising.

Looking ahead, we are still expecting an upgraded voice mode in the coming weeks. Hopefully, we will also hear more about Conway. For now, we will have to wait and see when those updates actually arrive.

Alt text

🔥 Anthropic adds usage reflection dashboard to Claude for all users - Anthropic introduces a reflection dashboard for Claude, letting users review activity trends and set quiet hours or break nudges via Settings.

Alt text

🔥 Anthropic brings Claude Cowork to web and mobile for Max users - Anthropic expands Claude Cowork to web and mobile, allowing users to start tasks on desktop and manage them anywhere.

Apple

ByteDance

Alt text

Exclusive: Early 30-second AI videos generated by Seedance 2.5 - ByteDance prepares to launch Seedance 2.5, its model for native 30-second video generation, with API access likely starting July 16 after delays.

Alt text

🔥 ByteDance debuts Seedream 5.0 Pro with advanced reasoning - ByteDance has unveiled Seedream 5.0 Pro, a multimodal image-creation model offering precise edits, multilingual support, and advanced features.

Cursor

Alt text

Cursor 3.11 adds side chats and agent transcript search - Cursor 3.11 adds Side Chats for parallel agent conversations and full transcript search, advanced agent controls, and improved project selection.

DeepSeek

Google

Things were very quiet on Google’s side this week. The company introduced custom URLs for AI Studio Build, allowing users to host their web-coded apps under an AI Studio domain. It is a neat addition and slightly resembles the experience Google offered around 2010, when users could build websites hosted under a Google domain. In a way, we have come full circle.

We also saw Gemini 3.5 Pro checkpoints being tested on LM Arena. They appear significantly stronger than the versions tested in June. However, it is still unclear whether Google plans to release anything soon or continue improving the model and delay the launch further.

Meta

Meta brought several surprises this week! First, the company released Muse Image and previewed its Muse Video model. This means Meta will no longer need to rely on Midjourney, which, in my opinion, is a very good move. Muse Image performed remarkably well, especially considering the scale at which Meta was able to expand access to the model. For many users, it represents a substantial upgrade.

However, some of the newly released functionality was later removed from Instagram. For a brief period, it was possible to generate an image based on any Instagram account, but that is no longer available, for obvious reasons. Overall, it was a strong announcement. We still do not know when the video model will arrive, but hopefully the release will happen soon.

Meta also launched Muse Spark 1.1, and the upgrade was surprisingly good. Muse Spark 1 was released only recently, so the arrival of version 1.1 suggests that Meta may be capable of delivering regular model improvements. Consistent upgrades also appear to be part of the company’s broader strategy. This was the signal I had been waiting for before saying that Meta was truly back in the AI race. I am now confident that it is. We will be following the company closely because, with its resources, Meta can easily become a serious part of the competition.

Alt text

🔥 Meta debuts Muse Spark 1.1 model and opens API for developers - Meta introduces Muse Spark 1.1 with a public API preview, offering advanced AI agent capabilities and improved multimodal workflows.

Alt text

🔥 Meta launches Muse Image across its apps and previews Muse Video - Meta AI introduces Muse Image, an image-generation model now live in Meta AI apps and Instagram Stories, offering prompt-based visual creation.

Mistral

Alt text

Mistral unveils Robostral model for robot navigation with RGB camera - Robostral has unveiled Navigate, an 8B-parameter vision-language model guiding robots through real-world spaces using only a single RGB camera.

OpenAI

Things were massive on OpenAI’s side this week. We got GPT-5.6, ChatGPT Work, and GPT Live 1, the upgraded bidirectional voice mode that has now been rolled out to everyone. Overall, it represents a significant improvement.

I spent several hours in the car this weekend and was able to spend that entire time talking to ChatGPT’s voice mode. The experience is substantially better. Interruptions are almost nonexistent, and the interaction feels genuinely conversational. Until now, only Sasame had offered this capability, but the technology is now available to everyone through ChatGPT. The biggest remaining gap is tool use. Voice mode can display some basic widgets based on information retrieved through search, but it cannot yet operate apps, plugins, skills, or connectors. If OpenAI manages to add these capabilities, voice mode could become considerably more useful.

GPT-5.6 is also performing very well. People are praising its ability to serve as a strong daily driver. From a cost perspective, I think this is one of the most important areas for OpenAI to target. Users can access some of the latest intelligence advancements at a reasonable price, without rapidly exhausting their limits with models such as Terra. Luna is also performing very well.

The long-awaited merger between ChatGPT and Codex has also happened. We now have a new desktop app built on top of Codex that includes ChatGPT Work and its related functionality. It will be interesting to see how the product develops from here. We will continue following all updates around these apps closely, so stay tuned!

Alt text

OpenAI AMA: Reddit summary about new ChatGPT and GPT-5.6 - OpenAI’s Codex now serves over five million weekly users, with new guidance for GPT-5.6 models and workflow tips shared during a Reddit AMA.

Alt text

🔥 OpenAI launches ChatGPT Work for Pro, Enterprise, and Edu plans - OpenAI introduces ChatGPT Work, allowing users to execute projects and automate workflows across apps and web tools on all platforms.

Alt text

🔥 OpenAI launches GPT-5.6 Sol, Terra, and Luna on apps and API - OpenAI launches GPT-5.6 with Sol, Terra, and Luna models for ChatGPT, Work, Codex, and API, offering higher intelligence.

Alt text

OpenAI set to launch ChatGPT Work upgrades today - OpenAI is set to introduce ChatGPT Work with features for team collaboration, workspace resets, and custom workplace assistants.

Alt text

🔥 OpenAI rolls out GPT-Live voice for ChatGPT on web and mobile - OpenAI introduces GPT-Live 1 for ChatGPT Voice, supporting real-time conversations with a global rollout to Go, Plus, Pro, and Free users.

OpenClaw

Perplexity

Prime Intellect

Tencent

xAI

It was also a massive week for xAI. The company released Grok 4.5, which can now arguably be considered one of the top three models in terms of intelligence.

The model is available for free to anyone with an X or Grok account through Grok Build, making this a good opportunity for users to try it for themselves.

Grok 4.5 is xAI’s first 1.5-trillion-parameter model and reportedly also relies on data from Cursor. We will likely see it improve further over the coming months. The bigger question is whether xAI can maintain a frequent release cadence and keep pace with the progress of other labs. At the moment, speed appears to be the main challenge, not only for xAI but for nearly every lab outside the top two.

Alt text

🔥 SpaceXAI launches Grok 4.5 model on Grok Build and APIs - SpaceXAI launches Grok 4.5, its coding-focused flagship model, now available in Grok Build, Cursor, and its API console.

Alt text

🔥 SpaceXAI launches 21 new Grok Voices on xAI Console - xAI rolls out 21 new Grok Voice flagship voices, boosting multilingual support and production-ready voice agent creation.

Alt text

SpaceXAI gearing up for upcoming Grok 4.5 release - References to Grok 4.5, built on a new V9 foundation, are surfacing on the Grok web, hinting at a public release for Heavy subscribers soon.

Zhipu AI

Happy testing! 🤖

P.S.

If you use Telegram, don't forget to join us there too! It is the fastest place to learn about new AI scoops 🗞️

P.S.S.

If you want to promote your AI tool in this newsletter, please check our Stan Store for Advertisers.