Inception's Diffusion LLM API Is Shutting Down
Sad to hear, but I got an email from hello@inceptionlabs.ai announcing the Mercury API that hosts their diffusion models is shutting down.
No blog post yet but I assume one is coming.
Here is the full email: ``` Hi,
We want to share an important update regarding the Mercury API. After careful consideration, we’ve made the decision to wind down the service. The Mercury API will remain fully available through December 28, 2026.
After this date, you will no longer have access to the service or any data submitted to the platform. We encourage you to retrieve any necessary data before then. If you have questions or need assistance with the transition, please reply to this email. We’re happy to help.
It’s been a privilege building Mercury alongside customers like you. Your feedback and ideas have been invaluable in shaping the platform, and we’re deeply grateful for your support and trust.
Warmly, The Inception Team ```
Also a shorter message on their discord from Yannis Miraoui: ``` Hi everyone, API access to all Mercury models will end on December 28, 2026. Mercury models will no longer be available from Nov 8 via OpenRouter.
Thank you for building with Mercury and for being part of this from the start! Please contact us if you have any additional questions. ```
"Which decision model should I use now?"
1. Jev - after Mercury Decide, it is the best value
2. Gemini Decisions - better calibrated, better accuracy, higher price. Excellent free tier, but they train on your prompts
3. Cloudflare Clef - clearly trained for internal use to be cheaper than 3p alternatives, but then sold to customers. primary advantage is larger context window
4. Laya - tiny context window. only good if you fine tune it / slap a LoRA on it. Given inputs that don't vary too highly from the SFT training sample, this can be the best option. Can be extremely fast. H100 can train a LoRA in minutes. Modal free credits are enough to train several Laya LoRAs.
I'm confused about this announcement. What does it actually mean? Will the Mercury models be available from somewhere else..? Is the Inception API different from the Mercury API? "Inception API" is what their marketing talks about
Are they getting acquired or something? I really like Decide, it did better than Jev for me, so I was going to use it pretty heavily in a few of my apps.
I found one report that Microsoft completed a transaction purchasing Inception Labs, but it makes no citation.
It's a bit fishy.
phew. what a shame, mercury 3 was a very good model for specific use cases. what are you all planning to switch to?
When I read that mail, I honestly thought this was some incredibly targeted phishing attempt or the like. Not having a blog post, nor anything on their other media channels is a choice to put it mildly.
Very disappointing, their models were so cheap and SO fast. Wonder if there's some type of merger or acquisition? And what will happen to the tech? imagine if they open sourced it all..
Really disappointed about this. I've been using Mercury 3 quite a bit and with great success. It is an odd announcement for sure. Makes me think acquisition of some sort. I'd say keep an eye out for future press releases regarding the company. This was basically a heads up to the people using the model to not rely on it long term.
.... will they open their diffusion models? their newest coding one for all its dumb has the wonderful property of admitting when they don understand stuff and proactively looking stuff up. i was really planning to try to use it more... i really hope they mike drop their ip at least then