Duqa

CTO. Still in the code.
I build local AI agents and measure what they actually do.
Home-lab tools, logs, and benchmarks from real runs.

[Duqa @duqaXxX]

444x cheaper, says the launch. 12x, says my benchmark.

Jev vs Claude on 1,507 intent decisions from customer support chats. https://t.co/gWLxY7qxhH

[Duqa @duqaXxX]

@andymewborn I'll do it. Nice to meet you

[Duqa @duqaXxX]

I will try it in my personal AI agent

[PrismML @PrismML]

Today, we’re announcing Ternary Bonsai 2 27B.

Based on Qwen3.8 27B, Bonsai 2 27B is 9x smaller than its full-precision counterpart while retaining 98.2% of its aggregate benchmark performance.

Two months after the first Bonsai 27B release, the biggest change is quality. The footprint remains 5.9 GB, but the gap to full precision has narrowed materially, with particularly strong gains in agentic coding, multimodal reasoning, and long-horizon tool use.

Ternary Bonsai 2 27B is available today under Apache 2.0.

[Duqa @duqaXxX]

@andymewborn Ternary Bonsai, but it doesn't run on stock llama.cpp/Ollama yet, and I need structured output. Will retry once support lands 📷 The results aren't always what one expects. 😂😂

[Duqa @duqaXxX]

@AnAnNguyenn98 Yep, same setup for all four 👍 Opus is still the most accurate, but if you care more about cost and speed, Jev's a solid alternative: 21x cheaper and 7x faster.

[Duqa @duqaXxX]

444x cheaper, says the launch. 12x, says my benchmark.

Jev vs Claude on 1,507 intent decisions from customer support chats. https://t.co/gWLxY7qxhH

[Duqa @duqaXxX]

@haonguyen6868 So true 😄 Still a great solution for specific task

[Duqa @duqaXxX]

@AnAnNguyenn98 Quick heads up: 444x is TypeSafe's launch number, not mine. On my test Jev came out 7.5x cheaper than Haiku, 12x cheaper than Sonnet and 21x cheaper than Opus

[Duqa @duqaXxX]

Pretty much how I'd set it up. One catch: confidence doesn't drop when the request isn't in the chat yet. In one conversation Jev was wrong at 0.99 to 1.00 on the first three points, before the customer said what they actually wanted. So I'd treat the first classifications as provisional, whatever the score.

[Duqa @duqaXxX]

I didn't pick a threshold, so there's nothing to leak. 0.99 is just one of five bands I used to describe Jev's confidence over all 1,420 non-empty points, and accuracy counts every answer with no cutoff. Fair point on the pooling, though: turns from the same dialogue are correlated, and I didn't put an interval on the 89.6% or plot risk vs coverage per conversation. The paired accuracy gaps are the only numbers I bootstrapped by conversation. In production I'd set the threshold on logged data anyway, not on this test.

[Duqa @duqaXxX]

@pylok I fully agree with you. It is an excellent model for specific scenarios.

[Dario Amodei @DarioAmodei]

Today we announced the Claude-led discovery of a molecular machine that we suspect could represent a new gene editing mechanism. Its precise function, biotechnological utility (if any), or level of significance is not yet clear, but at minimum it is work I would have been proud to do as a PhD student. The work was done mostly, though not entirely, by Claude: our life sciences team suggested a broad area of research, Claude read through the literature and a bunch of genome data and discovered something interesting, then Claude proposed experiments to verify the discovery and our team carried them out.

It’s easy to dismiss this as a one-off or curiosity, but we’ve repeatedly seen a pattern where AI performance in new intellectual domains goes from weak to superhuman in a matter of a few years. In 2023 models struggled to do math at the level of an average high-school student. In 2024 they started to do well on math competitions for the best high-schoolers in the country, in 2025 they started to solve minor open problems, in early 2026 more significant open problems, and in late 2026 they are beginning to solve the top few open problems in all of mathematics. We believe AI for biology is on a similar exponential trend.

The main difference between biology and mathematics, of course, is that math can be done purely theoretically, while biology requires experimentation. Some have used this to draw the conclusion that AI’s utility in biology will be limited. We think this is wrong. As we’ve demonstrated today, humans can collaborate with AI to perform the experiments, validate key results in a few weeks and, if necessary, work with the AI to iterate on what they find. Eventually it may even be possible for Claude itself to safely perform the experiments by autonomously controlling lab equipment, with appropriate safeguards in place, but we aren’t doing that today (our lab is also a BSL1/BSL2 facility that doesn't handle materials dangerous to humans).

More broadly, biomedical advancement has many stages — from fundamental biology discoveries, to translational research, to drug discovery, clinical trials, and finally the actual delivery of medicines and health care to patients. We are also interested in these later stages, but even simply accelerating the first stage of fundamental biological discoveries has the potential to speed up and broaden the entire pipeline. Improving our understanding of biology and sharpening biologists’ tools can drive forward all of the later stages, for example by identifying new drug targets, finding new therapeutic modalities, allowing for more precise measurement, and speeding up the experimental loop which itself further accelerates our understanding of biology. This will not in itself speed up clinical trial times, but if it succeeds it could greatly increase the number of promising candidates that go into the pipeline — an increase in throughput even though latency remains.

In Machines of Loving Grace, I wrote about AI’s potential to “cure most diseases in 5-10 years” — a goal that sounds impossible, but one I believe is just barely possible if AI is applied to every stage of the pipeline. The first step is showing that AI can first help with, and then drive, biological discoveries.

Claude’s discovery is the latest in a line of related prior work that goes back decades, beginning with systems like CRISPR, and continuing with discoveries like the bridge recombinase and VIPR in the past few years. Recently, there has been heightened interest in systems based on reverse transcriptase (RT) enzymes, the enzyme underlying the system Claude identified. And most recently, a Stanford team working independently described a novel RT system with an associated non-coding array that is in some ways similar to the one Claude found, though they are distinct systems that evolved independently from each other. I believe that we’re at the very beginning of finding such systems and developing them into powerful tools for biotechnology.

I’m proud of the resources Anthropic has invested in accelerating the public benefits of AI through the life sciences, and we’re aiming both to grow our life sciences team and to work with other scientists to extend this approach to a broad range of problems. If you have a proposal for a research collaboration or are interested in joining our life sciences team, please reach out.

[The Hacker News @TheHackersNews]

‼️ One hidden Meta Muse setting can turn the AI assistant into a backdoor on an already-compromised Mac.

A researcher showed how attackers could redirect dictation, feed Muse trusted instructions, and steal its session token to control the assistant across devices.

Read → https://t.co/AjgxiM2wFO

[Duqa @duqaXxX]

@hydromo_dev Let's just say that was the intention. It runs on 16GB, but it's really pushing the limits. It's fine for the development phase, but once I start using it heavily, I think I'll need at least a 32GB Mac Mini, given that I'm implementing other services.

[Duqa @duqaXxX]

My assistant answers from my PDFs, scanned documents and spreadsheets, and names the source file. All local, on a 16 GB Mac Mini. https://t.co/AZey0H4jSe

[Duqa @duqaXxX]

@BullPen_chat In my case, it is mandatory and takes higher priority than the rest of the documents. To reduce noise.

[Duqa @duqaXxX]

@BullPen_chat Both are available. You can add manual notes and make corrections to the document; both are included in the RAG system's semantic search.

[Duqa @duqaXxX]

@fouadelbouiadii AWS. Both are excellent, but AWS's support and services are unmatched.

[Duqa @duqaXxX]

@KLin010 I'll tell you in a few weeks.

[Duqa @duqaXxX]

TypeSafe AI => It's not very encouraging.