OpenAI unleashes big moves one after another!
Today (the 16th), a leaked internal Slack message revealed that ChatGPT is about to undergo an 'epic' performance overhaul.
Judging from the hardcore test data, the optimization of ChatGPT's frontend performance this time is so dramatic it's almost 'unbelievable':
Application loading speed directly skyrocketed by 94%, heap memory growth reduced by 87.8%, and overall memory footprint slashed by 41.2%.
Network requests plummeted by 98.2%, and conversation history loading was cut by a full 99.6%.
Almost simultaneously, Codex launched the GPT-5.6 'Multi-Agent v2' version.
The main Agent can now automatically delegate different subtasks to different models, and each sub-Agent can be individually configured with reasoning intensity.
OpenAI President Greg Brockman stated, 'Moving towards bidding farewell to manual model selection.'
It must be said, with this wave of 'two-pronged approach' from OpenAI—
It directly eliminates the tedious 'waiting' and 'choosing' for users, enabling ChatGPT to truly achieve a seamless and ultra-fast, fluid interactive experience.
ChatGPT Major Overhaul, Loading Speed Soars 94%
What many don't know is that this major performance upgrade for ChatGPT is actually a crucial battle after the departure of 'Programming King' Scott Gray.
When using ChatGPT, what people have always dreaded is that long conversations become slower and more sluggish, with the web page even crashing directly.
On the official OpenAI forum, someone even posted asking, 'Is the new Codex slower?', with many agreeing in the comments.
This time, Codex lead Andrew Ambrosino's shared internal Slack screenshot laid it all out:
What they used for internal testing was a 'monster-level' conversation with a full 741 rounds and a size of 231MB!
This might sound like an extreme case, but it's the norm in the Agent era.
In the past, conversations were chat histories, maybe a few dozen rounds at most; now, a slightly complex task involving reading code, running tests, making changes, and verifying again can easily reach hundreds of rounds.
It's this kind of high-intensity conversation that now opens in an average time reduced from 27.62 seconds to 1.66 seconds.
Total application memory growth was compressed from 1030.7 MiB to 606 MiB;
The network requests needed to open it once went from 894 down to 16; the number of session entries that needed loading dropped from 15,529 to 64.
This is a major frontend optimization by OpenAI targeting ChatGPT's ultra-long conversations.
In two words, smooth.
What difference will it actually make?
Simply put, those chat logs spanning months will open much faster.
Developers who frequently run heavy Codex sessions with hundreds of tool calls will find the interface noticeably snappier, and their device's memory usage will drop accordingly.
Switching back to an ultra-long old conversation will no longer feel like 'loading a save file from a nearly-dead PS3'.
The clever part is that ChatGPT no longer needs to load and render the entire history just because you opened a conversation.
It can store most of the content and only load the small portion people actually need at the moment.
The same conversation, but with a huge amount of unnecessary work eliminated.
For those accustomed to living in ultra-long AI conversations, this is the kind of seemingly boring infrastructure upgrade that can dramatically improve the product experience.
Multi-Agent v2 Fully Launched, No Need to Manually Select GPTs Anymore
Just this week, OpenAI quietly updated ChatGPT.
It wasn't until today that OpenAI engineer Eric Provencher officially announced: Multi-Agent v2 is now fully live.
The focus is on intelligent task division.
The main Agent can automatically delegate subtasks to any supported model, including Luna, with each sub-Agent supporting independent reasoning intensity settings.
Now the hand ChatGPT/Codex holds looks like this—
GPT-5.6 Sol: The strongest, for complex Agentic coding
GPT-5.6 Terra: The daily programming workhorse
GPT-5.6 Luna: The fastest, cheapest
Daybreak: Specialized for cybersecurity
GPT-5.5: Complex programming, research, general tasks
Actually, three weeks ago, in Codex's model list, Sol and Terra were marked as multi_agent_v2, while the cheapest Luna was marked as v1.
The result was that when the main Agent tried to delegate work to Luna, the system would directly reply with 'unknown model'.
On GitHub, developers opened at least two issues about this.
There's a post on the official OpenAI forum titled 'Give us Luna back', where the poster said they previously used a hook to force using Luna with amazing results, and now that path is blocked.
Previously, people needed to manually choose models; in the future, the Agent system will automatically split tasks, execute them in parallel, and summarize results, turning models into internal computational resources.
For complex tasks, only 20% of the steps might need the strongest model, while the rest can be handed off to cheaper models.
This way, reasoning costs can be significantly reduced, accelerating ChatGPT's transformation from a chat tool to a workflow platform.
ChatGPT Evolves On the Spot, The Fully Automated Monster is Here
The frontend has cleared away the 'historical baggage', and the backend has enabled 'intelligent distribution'.
This underlying 'one-two punch' from OpenAI not only makes ChatGPT completely bid farewell to the era of lag but also announces to the entire industry:
Large models are accelerating their evolution from 'chat tools' to true 'fully automated workflow platforms'.
No more worrying about which model to choose to balance costs, no more painfully waiting for monster-level hundreds-of-rounds conversations to slowly unfold.
Just throw the complex task over, and leave the automatic decomposition, model scheduling, and ultra-fast rendering to the system.
The era of ultimate smoothness for Agents has truly arrived.
Faced with such a 'seamless' and powerful ChatGPT, is your productivity ready for takeoff?
References:
https://x.com/Ananth7e/status/2088490421676863782?s=20
https://x.com/ajambrosino/status/2088401536057827344?s=20
https://x.com/gdb/status/2088658133971509640
This article is from WeChat Official Account 'New Zhiyuan', author: Taozi
