Last night @PrismML shipped Bonsai 2 and @darkbloomai is the first model provider to support it. We're offering the first 250 users 100m free tokens test it out. At 9x smaller than Qwen 3.8 with 98.2% of the benchmark performance. It outperforms Opus 4.6 and 5.6 Luna and its the first model that can run on the lightest weight personal MacBooks in our Darkbloom fleet. The race for bigger centralized models misses half the point. Models like Bonsai are how AI becomes ubiquitous and Darkbloom ensures the spoils of the inference economy are accessible to everyone.

> be @darkbloomai >gather everyday people's Macs >connect them to a grid >create a private "local AI as a service" >optimize models like @googlegemma @Alibaba_Qwen and @nvidia nemotron with your sister team @yukonresearch >run them 2-3x faster and 50% cheaper by stripping out datacenter margins >partner with @OpenRouter for distribution >pass earnings along to passionate, growing local AI community that offer up their spare compute. rinse. lather. repeat

what a week to publish this paper has been a live demo of what open, multiplayer research collectives can achieve together. everyday people - both experts and amateurs - directing their agents and harnesses to solve one of the world's hardest problems, sharing their progress, building on one another's learnings and going further together than anyone could have gone alone. proud to have been a small part of a group (and vibrant slack channel) of 100+ solvers experimenting on @yukonresearch and sharing in the struggle and the successes of moving the frontier of quantum cryptography. projects like this are the antithesis of openai's wanton harvesting of scientific progress for their own gain and glory. its what open, scientific progress SHOULD look like. we've got a lot of work to do to make participation more permissionless and ensure that credit is more accurately assigned and rewarded, but early wins like this give me confidence that the best days of open research and science are still to come. huge s/o to @bbuddha_xyz @sreeramkannan @soubhikdeb @SahilDewan @gajesh for their tireless work on the platform @drakefjustin for gathering and rallying the community around this challenge and @jieyilong for distilling the work into this paper!

Good piece on @darkbloomai from @TheRegister this morning. "@eigenlabs, a five-year-old tech biz based in Seattle, Washington, estimates that owners of Apple Silicon hardware can earn $120 to $200 per month on average by selling idle compute power for AI inference. Whatever the payout, that's money going toward people running open weight models on personal computers and not frontier labs with data center debt that are struggling to attract customers to premium models." Full story here:

such a great, well reasoned piece by @tarunchitra i spent years on the OSS team @Azure building commercial businesses around open software (and OSS hated microsoft pre-satya) linux, kubernetes and oss dbs are now billion dollar businesses in azure. vs code and .net are now open. to win you had to know the chessboard and build bridges to communities at the right layer of the stack. tarun's laying out the map for open weights. tarun is spot on that OSS incentives followed from having more upfront cost to develop/maintain vs. ongoing cost to use and run the software. open weights will have its own business models given its more compute/capital intensive both to develop (training) and use (inference). we think a lot about this for @darkbloomai in particular, which doesn't neatly fit in any of his four horseman buckets. we're sort of like a rebater that can offer a structural discount vs. model builder rates because our underlying machines are idle/sunk cost consumer hardware. we're also kind of a specialist, except our HW arrangement doesn't justify a premium, its what lets us offer cheaper inference. early days, but as bertrand competition drives provider margins down to his Souq endgame of OpEx floor compression, offering access to machines with near zero reservation price may be the winning value prop.

Love this episode Our friends @a16z noted earlier this year that inference is the "COGS of intelligence" Here @sreeramkannan goes further, calling out that "as intelligence becomes a more important input into everything we build, who gets to participate in creating it matters enormously." @darkbloomai lets any Mac owner participate in the post-AGI inference economy and @yukonresearch aggregates the participation of hundreds of harnesses to move research and science forward together, in public. Closed intelligence is not the only way. Power concentration my a few labs is not predestined. Open innovation is cryptos great gift to AI and will be how we preserve and expand individual agency in a post-AGI world.

Public autoresearch does it again Qwen 3.8 27B now runs 3x faster on Apple Silicon. Over the weekend @gajesh @0xkydo @soubhikdeb @bbuddha_xyz and the team at @yukonresearch dropped a new challenge, this time targeting Qwen. As we learn more about open autoresearch, we're offering $10k each week to solvers who help move the benchmarks on these open research challenges. Push the frontier at

Yukon is live. @poolsideai 's open-weight model > 2.6x faster @Ethereum's post-quantum scaling > up 3.5x @Lighter_xyz 's prover > 9.5x faster Researchers at @Princeton and @UCBerkeley made progress on 150+ science and math problems