It’s true. Ternary Bonsai 2 is great as an OS assistant and for small coding tasks, with excellent instruction following and long-horizon coherence. Replaced Gemma 4 as my local daily driver. 262k ctx, Q8 KV, 3GB VRAM free on a 4090, ~130 tok/s with DFlash enabled. Incredible.

👀

The long coinbase maturity soft fork is much cleverer than it looks at first glance. It doesn’t try to identify or ban mercenary BLAKE2b hash. Instead, it attacks the economic machinery that allows immature SV1 pools to rapidly aggregate opportunistic hash and turn it into liquid payouts. At block 973,440, newly mined coinbases become subject to a 6,480-block maturity until block 979,920. But because the enforcement window itself is exactly 6,480 blocks long, none of those rewards actually mature while the rule is active. They all hit the same cliff: effectively, every new mining reward is frozen until 979,920, when consensus returns to the normal 100-block maturity. That is brutal for an immature SV1 pool. If its business model relies on spending coinbases around the normal 100-block maturity to pay transient hashers, that flow simply stops working. Spend one of those covered coinbases too early and upgraded nodes reject it; mine a block containing that spend and those nodes reject the entire block. The alternative is for the pool to keep paying miners out of its own already-mature reserves while 45 days of newly earned block rewards accumulate frozen behind it. For a thinly capitalized pool, that can be existential. And the more opportunistic hash it attracts, the larger the liquidity hole becomes. The mechanism turns the very thing that makes SV1 dangerous (being able to rapidly aggregate huge amounts of mercenary hash) into a massive capital requirement. DATUM with direct generated payouts, such as TIDES, fits the model much more naturally. The miners' shares are placed directly into the coinbase when the block is created. Those individual outputs still have to mature, but there is no centralized pool treasury that first receives the subsidy and then needs to spend it later to distribute rewards. The maturity burden follows the actual miners rather than forcing a pool to finance an army of transient hashers. Another clever part: there is no miner or node signaling threshold. This is a height-based flag-day soft fork. Nodes running the new version begin enforcing it automatically at 973,440 regardless of how many other nodes signal support. Nodes that refuse to upgrade are not given veto power over activation. As long as miners follow the stricter rule, they can continue following the same chain. But if an old-rule SV1 pool or miner tries to spend one of these coinbases after only ~100 blocks and mines that spend, the enforcing network rejects the block outright. Any non-upgraded nodes that accept and build on it have followed that miner onto an incompatible fork. That is another benefit of the design: unaligned miners do not get to drag enforcing nodes back to the old rules. If they insist on monetizing rewards before the new maturity permits it, they effectively fork themselves, and any old nodes willing to follow them, off the enforcing network. That's the elegance of it: the network doesn't need to know which ASICs are “mercenary.” It changes the incentives underneath them. Opportunistic hash can still point at XBT, but the thinly capitalized SV1 pools that make that hash liquid, convenient and scalable suddenly need enough reserves to finance weeks of payouts, or make their miners wait.