AI tool comparison
Bonsai-8B vs MemPalace
Which one should you ship with? Here is the side-by-side panel verdict, pricing read, reviewer split, and community vote comparison.
Infrastructure
Bonsai-8B
A true 1-bit 8B LLM that fits in 1.15 GB — runs on your iPhone
75%
Panel ship
—
Community
Free
Entry
Bonsai-8B is PrismML's latest model in their BitNet-inspired lineage — an 8.2B parameter language model that has been quantized end-to-end to true 1-bit precision (weights stored as -1 or +1), compressing the entire model to just 1.15 GB. That's roughly 12-14x smaller than a standard FP16 equivalent. Unlike post-training quantization hacks that lose substantial quality, PrismML trained Bonsai-8B with 1-bit arithmetic baked into the forward pass from the start. Benchmark results are competitive for the size class: 63.8 on MMLU, 72.1 on HellaSwag, and 54.2 on GSM8K — while running at 131 tokens/sec on an M4 Pro MacBook and 44 tokens/sec on an iPhone 17 Pro Max. That makes it the fastest locally-runnable 8B model in its weight class on Apple Silicon. The MLX-optimized weights are available on Hugging Face today under Apache 2.0. The significance goes beyond benchmarks. Getting a capable open-weight model to run at interactive speeds on consumer hardware — with no API key, no GPU, no cloud dependency — is a meaningful step toward truly private, offline AI. This follows PrismML's earlier "Ternary Bonsai" (1.58-bit) but represents a cleaner binary architecture that's easier to accelerate on custom silicon.
AI Memory & Context
MemPalace
Hierarchical cross-session AI memory — viral, controversial, open source
25%
Panel ship
—
Community
Free
Entry
MemPalace is an open-source persistent memory system for AI agents that organizes memories hierarchically — people and projects become "wings", topics become "rooms" — enabling scoped semantic retrieval rather than flat vector search. It claims 96.6% on LongMemEval and a 170-token overhead per session. MIT licensed, self-hosted. The project went viral almost instantly after actress and director Milla Jovovich pushed it to GitHub, claiming she built it with Claude Code alongside engineer Ben Sigman. The "palace" metaphor maps well to how humans naturally organize associative memory, and the architectural idea of scoped context windows (retrieve only the relevant "room") is legitimately interesting for long-running agent sessions. The controversy: GitHub issue #214 exposed that the headline benchmark measures ChromaDB's default embeddings, not the palace structure itself. The README was updated to walk back the "100% accuracy" claim. A pump-and-dump crypto token ($PALACE) also appeared within 24 hours of the GitHub push. The underlying memory architecture has real merit — the noise-to-signal ratio is just high right now.
Reviewer scorecard
“131 tokens/sec on M4 Pro at 1.15 GB is genuinely impressive — I can embed this in a macOS app without any cloud dependency, no rate limits, no privacy concerns. The Apache 2.0 license means I can ship commercial products on top of it. This is the edge AI story I've been waiting for.”
“The hierarchical memory concept is sound — scoped retrieval beats flat vector search for agents with complex long-term context. But the benchmark controversy (measuring ChromaDB embeddings, not the palace structure) makes it hard to trust the claims right now. Wait for independent replication and a clean README before building on this.”
“63.8 on MMLU is respectable but it's still noticeably behind mid-range cloud models on reasoning tasks. The GSM8K score of 54.2 means it'll fumble multi-step math that users expect to just work. Until 1-bit gets to 70B scale, it's a neat demo that falls short in production use cases where quality matters.”
“Celebrity open-source drop, inflated benchmarks, and a crypto token in under 24 hours — this is the trifecta of GitHub hype. The tech might be fine, but you can't evaluate it through the noise. Issue #214 alone should give any serious developer pause. Let the dust settle.”
“The trajectory here is what matters: 1-bit models are getting faster to train and competitive faster than expected. When custom Apple Neural Engine kernels land for BitNet-style weights, we'll see 200+ tokens/sec on a phone. Bonsai-8B is the proof-of-concept that makes that future feel real.”
“Strip away the celebrity drama and the palace memory metaphor is genuinely compelling. Agents that organize knowledge spatially — with room-level context scoping — are a step toward more human-like associative recall. The 23k star viral moment also signals serious latent demand for better AI memory primitives. Someone will clean this up and it'll matter.”
“I've been looking for something I can embed in a creative writing or brainstorming app that doesn't require an internet connection. At 44 tokens/sec on iPhone, Bonsai-8B is finally fast enough to not break the creative flow. The 'no account required' angle is a genuine selling point for privacy-conscious users.”
“The palace metaphor is beautiful UX-conceptually — I love the idea of 'walking' an AI through rooms of context. But the crypto token association makes me not want my name near this project right now. If the tech gets validated independently, I'm interested. For now, too risky.”
Weekly AI Tool Verdicts
Get the next comparison in your inbox
New AI tools ship daily. We compare them before you waste an afternoon.