← The Polylog AI Intelligence Brief
Morning Edition · Sunday, September 20, 2026Published at 2:20 AM EDT · New York
PrismML Compresses Qwen3.8-27B to 5.9 Gigabytes While Keeping 98.2% of Its Benchmark Average
Ternary weights cut the model to 1.76 effective bits, but long-horizon agent tasks degrade sharply, with Terminal-Bench 2.1 falling to 52.8 from 69.7.
PrismML, a startup founded by a team out of Caltech, released Ternary Bonsai 2 27B, a compressed version of Alibaba's Qwen3.8-27B. Each weight is stored as one of three values, minus one, zero or plus one, with FP16 group-wise scaling, giving 1.76 effective bits per weight. The full-precision reference occupies 53.81 gigabytes. The compressed model occupies 5.93 gigabytes, about nine times smaller, and the weights carry an Apache 2.0 license.
Across a 20-benchmark suite covering reasoning, math, coding, instruction following, vision and agentic tool use, PrismML reports an aggregate score of 83.9 against 85.4 for the original, or 98.2% retention. The Russian-language channel AI ML Big Data carried the same figures.
The aggregate number hides where the loss concentrates. PrismML's own breakdown puts math and coding at parity, vision retention at 96.3%, and knowledge and reasoning at 96.9%. Long-horizon agent work falls much further. On Terminal-Bench 2.1 the compressed model scores 52.8 against 69.7, and on SWE-bench Verified it scores 60.8 against 80.6. Those are the two evaluations that most closely track what an agent framework actually asks a model to do over many turns, and a drop of roughly 20 points on each is not a rounding error.
All of these numbers come from the vendor. No independent reproduction has been published. The compression itself is verifiable by anyone who downloads the weights, which run on a 16-gigabyte laptop or a single 24-gigabyte graphics processor, so the claim is unusually easy to check.
What this means
Aggregate retention scores are the wrong metric for buyers building agents. A model that holds parity on single-turn math and coding while losing 20 points on multi-turn terminal and repository tasks is suitable for local chat, classification and retrieval, and unsuitable for unattended agent loops, and a single headline percentage hides that split. The gainers are laptop and workstation silicon vendors and anyone serving latency-sensitive inference on-premises, since a 27-billion-parameter class model now fits in consumer memory. The losers are API vendors whose lowest tiers exist to serve exactly the summarize-classify-extract workloads that survive ternary compression intact.
What to watch
Part of a tracked trend
AI Inference Shifts to Consumer Devices
Over the next 3-6 months, smaller efficient architectures and inference-cost optimizations push capable AI off the cloud and onto laptops, phones, and mobile NPUs.
Townsquare
Start a discussion in Townsquare.
More from this edition
- Anthropic Weighs a New Claude Model as OpenAI's Astra Overtakes It in Tracked Enterprise Spending
- A Chatbot Invented a Nuclear Cargo and the US Military Prepared to Board a Chinese Ship
- Researcher Factors RSA-896 With Claude's Help, Advancing the Public Factoring Record 34 Bits in 16 Days
- Anthropic Models Capital's Income Share Rising to 54.8% by 2030 in Its Most Extreme AI Scenario
- Anthropic Says It Disrupted Claude Misuse Across Seven Harm Areas, Including a Russia-Linked Espionage Campaign
- Claude Code Projects Now Runs Parallel Cloud Agents, Each on Its Own Branch
- Anthropic Publishes Three Self-Measured Development Metrics, Reporting Claude Leads 26% of Its AI Research Work
Comments
0No comments yet.