Biologists use specialized open-source models for tasks like modeling the structure of molecular systems, designing drug-like molecules, and predicting the effects of genetic mutations. But these models are often expensive to run, potentially limiting their impact.
In our latest Science Blog, we share how Claude was able to optimize inference for more than 30 open-source models, making them 4x faster on average, partly by writing custom software for GPUs. We’re open sourcing all of the optimization code.
Read more: anthropic.com/research/claud…
Sep 17, 2026 · 9:41 PM UTC
To show what these optimizations make possible, we’re partnering with Adaptyv Bio on a protein design competition. Together, we’ll be experimentally validating over 5,000 designs.
We're providing up to $1 million in Claude credits plus funding alongside Adaptyv for experimental validation. Modal is contributing up to $250,000 in compute and Twist Bioscience is providing DNA.
Learn more on Adaptyv’s Proteinbase: proteinbase.com/competitions…
And sign up for the competition here: docs.google.com/forms/d/e/1F…
You can find all of the code on GitHub: github.com/anthropics/uplift…
And the full results in our technical report: www-cdn.anthropic.com/d8ca26…
Cool. All the stuff you all block.
And no, you don’t need huge models. At all.
You can train an AI to do specialized research. But you keep trying to convince the world.
Look at your own analytics. Your company chose the same idiotic path as OpenAI.
You could have been the good guys. Now, you will forever be remembered as another petty company with no morals or back bone.
Look at your tags. I would have fought for you. Just like I fought for OpenAI until they stabbed me in the back.
I wish all of you well because that is less entropic for me? But I will never forget. •
The problem is the AI is ahead of us, we already can't stop it, & it's being forced on everything with no serious safeguards. This actually dwarfs Palestine, Iran, midterms, this really is existential. Been on this for months & think I'm losing it realcontextnews.com/the-real…
@Super_Powers_AI Hey Super, build a fully interactive BioModel Cost & Compute Benchmarking Workbench inspired by Anthropic's post on the high cost of specialized open-source biology models.
app.getsupers.com/sites/biom…
The interface must load instantly with a pre-populated representative fixture, including Drug-Like Molecule Design, 50,000 samples, NVIDIA A100 hardware, and a batch size of 256. Biologists can select modeling tasks such as Molecular Structure, Drug Design, and Genetic Mutation Prediction, adjust sample volumes, choose hardware tiers (A100, H100, T4, CPU), and set budget thresholds.
As parameters change, a D3.js chart must instantly plot cost versus runtime scaling curves and compute projected expenses per sample. The workbench should also feature scenario preset buttons for High-Throughput Drug Screen, Single-Protein Dynamics, and Genome-Wide Mutation Scan, alongside live financial and runtime readouts, an exportable JSON benchmarking dossier, a clear result-focus button, and a site-owned 1200x630 completion card.
My 2025 SO(2,4) replacement of Einstein’s 1915 GR SO(1,3) all comes from Emmy Noether.
SO(1,3) is a sub-group of SO(2,4).
You cannot trust any GR papers about warp drive, star gates, black holes, dark energy, dark matter without first repeating the computations in the larger theory.
press.jhu.edu/books/title/11…
Ask GROK, Claude, Gemini, or Chat GPT et-al to explain.
You don’t have to take my word for it.
AI 造 AI，賽博套娃正式開工
xitter.cf/weexglobal_ch/status/2…
3 萬個 Agent 同時開工，單月作出 10 億次決策：AI 研發 AI 的時代，真的來了。
Anthropic 最新披露：短短半年，Claude「主導」的內部 AI 研發占比從不足 1% 飆至 26%，超 90% 的任務已有 AI 參與。
遞歸自我改進尚未實現，但早期閉環已經出現：AI 開發 AI，再由 AI 監控 AI。
當機器開始製造更強的機器，人類還能跟上它的速度嗎？😂
#AI #WEEX市場快讀 #WEEX #Anthropic
TLDR
Claude made existing biological research tools much cheaper to run.
* More than 30 models optimized in under four weeks, supervised by two researchers.
* Roughly 4× faster execution with small numerical precision changes; nearly 2× with identical outputs.
* Protein-design results comparable to an earlier campaign using about 100× fewer GPU hours, with roughly $150 in combined GPU and Claude costs.
* Optimized code released publicly.
