Yesterday marked one month anniversary of @DarkbloomAI moving to paid tier on OpenRouter. We started with 4B tokens a day and today we are excited to be serving 20B+ tokens a day across 8 models. We have grown our provider base to 1.1K+ active at all times and plan to expand it much higher as we expand the model base and more powerful M5 chips are getting shipped over next few weeks. You can now run Darkbloom on any Macs (including corp) without our MDM. Last week, we piloted our new higher grade privacy model which allows us to bind machines to stronger trust guarantees leading to Apple's certificates. You must be on MacOS 27 to be part of this and you will be soon required to be on OS 27+. During last 2 weeks, we have shipped a bunch of features to increase the QoL for our providers which includes Global Payouts, better routing, MLX(.)Fast inference engine improvements etc. Thank you everyone for being early to our network and supporting us through all the journey. I also want to thank the OpenRouter team (@pingToven, @shashankgoyal95, Eddie, Will, @OpenRouterMatt) for making this happen and continuing to support us through adding new models. This week we're at our company offsite + retreat planning through all the things we need to ship in the coming weeks and months. We will share a much detailed update after this. If you are on Darkbloom, make sure you're part of the Slack community.

Darkbloom is the first model provider to support Ternary Bonsai 2 27B -- concentrated intelligence that fits on your phone. Try it now: First 250 users get 100 million free tokens; PrismML's new flagship model: - a ternary compression of Qwen3.8 27B at 2bits per parameter. - 8.5 GB total, 5x smaller than original - keeps 98.2% of the Qwen's FP16 benchmark performance. - 75% cheaper than Qwen 27B. Qwen3.8 27B already operates comparably with Opus 4.6 and 5.6 Luna on certain tasks. This one does it in the memory of a phone. 262K context, image input, Apache 2.0. From our first run on the network, on a single M5 Max with no caching: - 35 tok/s decode at 1K context, - 31 tok/s at 10K, - 19 tok/s at 50K. But we expect more performance gain coming in a few weeks! That's a full 27B reasoning model running comfortably on any Mac. 1,000+ Macs are serving on Darkbloom right now. Go try it out!! Thank you to @BabakHassibi @SahinLale @HessianFree @rsadri_ml @tushar_bans @evaninwords and the whole PrismML team. This is exactly the kind of model Darkbloom was built for. You can read the essay by Bonsai on Why Local AI Matters:

Together, the MLX(.)fast community has made Qwen 3.8 Flash nearly 2x faster on Apple Silicon! We're ready to bring it to @DarkbloomAI: an open network of local Mac machines providing inference to the world. One thing remains: the community flagged that its license requires a separate agreement for commercial model serving, so we're holding the launch until that's in place. We believe this is a great opportunity for the local community: one where we make Qwen models faster and more accessible, and the people running them share in the value they create. .@Alibaba_Qwen @QwenDevs, we'd love to work together on this. If anyone else knows someone we can talk to, we'd love to have that conversation. Let's make Qwen 3.8 Flash on Darkbloom a reality!

If you can coordinate people in masses flexibly, you can move mountains. We brought together thousands of people from diverse backgrounds (professors, engineers, cryptographers, landscape designer, 13yo high schooler, farmer, marketer) to solve this one problem. All our past work over 4 yrs revolves around that and two words: Open Innovation. We did that for compute (@DarkbloomAI) and capital before. Intelligence was the hardest format to crack and there have been different mechanisms. ECDSA(.)Fail was the first time where it worked at a huge scale. Thousands participated, hundreds got their contributions in. The format was pretty simple: have a hill-climbable goal and verifier that scores your work. AI played is important role in this - I’m the top of the leaderboard in terms of % of contributions yet I don’t know anything about quantum computing, barely about cryptography and high school dropout level math. I just knew how to build good harnesses (pre Astra, Fable etc) - Opus 4.6 is what we used mostly. The result of bringing everyone towards to solve a problem is the most exciting thing in world. We are doing that with Darkbloom and @YukonResearch (evolved and multi challenge scalable version of ECDSA Fail). We started with cryptography, did inference optimizations and we will continue to solve the hardest problems for humanity together. The beauty of living in the era of abundance. - This is also my first time on ArXiv. Thank you @JeiyiLong for writing this paper.

Darkbloom: Day 15 We have been locked in for last one week; It has been very exciting and we shipped a lot of improvements to the system. Metrics: - 16.3B tokens served: New ATH & 60% higher than last day. - $420K Network ARR (4x in 2 weeks); - 99.96% uptime (very happy considering we're a distributed network). - ~15-17% utilization at peak; median: 13% utilization. Product: - One big win: we take up 30% of traffic for Gemma 4 26B - We cut latency by ~1.5 second. We plan to ship few more features that would reduce latency by 30-90%. This will include cache-aware routing. - We shipped new routing and system optimizations which were key reason for us to cut down 429s. More such to come. - We will be adding 3-4 more models in the coming week. Excited for @Spangler3000 to be helping us with this! - We conducted MLX(.)Fast which has improved Gemma 4's performance by 2.5x -- the improvements are in process of being imported by @TheDavidTai. - We are exploring payouts solutions such as Stripe Global Payouts to add 67 more countries. - Astra has redesigned our Console UI (this is AGI, ngl). We will also be starting up our efforts for the MacOS app for better user experience. - We have also been looking more deep into Autopilot where we load up models according to the traffic and other factors to maximise the provider earnings. - Our new landing page is almost ready and coming up with one suprise ;) -- huge kudus to @0xkydo and Estella. - Also huge thanks to @iLLRiPHiTTeR for helping us while onboarding new models on OpenRouter. We were able to find out minor issues in our inference engine. - We are also working through engineering practices and cleaning up code. Thank you everyone for being part of this journey and reading along this update. We build these systems to bring people together to contribute to this AI economy. As always, it has been rewarding to keep doing what we do. Until next time.

exciting 72 hours. thank you everyone for this overwhelming amazing response around darkbloom!! touched some grass today. we’re working through the issues with the great inflow of users and more powerful machines. please bare with us. we are also working towards adding more models and autopilot.

Darkbloom is 4 month old with early PMF signals (~100k yearly run rate). We are scaling and are looking for extraordinary talent. If you are someone who is excited: 1. To get shit done 2. To fail and learn new things 3. About distributed AI & systems Send me or @gajesh a DM with your proof of work. Engineers and operators are both welcome!

We are looking to hire an engineer and operator who can team up with me to accelerate the shipping velocity of Darkbloom. You will be working on the frontier of open distributed network on different systems (networking, inference, sandboxes, etc). Darkbloom is a shot for everyday person to not get left off from this compute gold rush. Reqs: good problem solver, deep divergent thinker and highly passionate; Fight on.