Open sourceHugging FaceConfirmed
Story 9 of 11 in the Thursday 22 October editionHugging Face releases Carbon-X, a 14.8B DNA-and-English language model
Built from Qwen3-14B with a hybrid BPE and 6-mer DNA tokenizer over a 40,960-token window, Carbon-X matches GPT-6.1 Sol across five genomics benchmarks at a fraction of the inference cost.
Edited by Ashutosh Shrivastava
22 Oct 2026, 23:55 IST · 1 min read
Hugging Face Biology Research, working with TIGEM, the University of Naples Federico II and Zhongguancun Academy, released Carbon-X (HuggingFaceBio/Carbon-X, Apache-2.0 licence), a 14.8B-parameter model adapted from Qwen3-14B with a hybrid English BPE and 6-mer DNA tokenizer trained on 302B tokens. Across five genomics benchmarks with thinking disabled, Carbon-X averages 66.7%, 27 points above Qwen3-14B and within 0.8 points of GPT-6.1 Sol (67.5%) at roughly 1/1,800 of the per-problem cost.
Sources
- PrimaryHugging Face: HuggingFaceBio/Carbon-X model cardhuggingface.co/HuggingFaceBio
- On XClem Delangue (@ClementDelangue) on Xx.com
The rest of the day
Read the whole day
01InfraAnthropic
Anthropic-managed Claude Platform becomes generally available on Google Cloud
Google Cloud customers can now access the full Anthropic-managed Claude Platform, including Managed Agents, Agent Skills and zero-day model launches, while authenticating with Google Cloud IAM and retiring Google Cloud spend commitments.
02Coding agentGoogle
Google releases open-source Gemini API CLI built for coding agents
Published on the Google AI Studio blog on 19 October 2026 and shared on 22 October 2026, the single-binary Rust tool gives coding agents and developers progressive terminal access to Gemini text, image, video, audio, embedding and research APIs without MCP schema overhead.
03CreativeIdeogram
Ideogram launches Ideogram Vector for native text-to-SVG and raster vectorisation
Ideogram’s first native vector model (ideogram-vector-1) generates editable SVG files with clean bezier paths, smooth gradients and sharp typography directly from text prompts or existing raster images in the web app and API.
04ProductOpenAI
OpenAI makes the Decisions API generally available for low-latency classifiers and rubrics
Powered by gpt-6-luna, POST /v1/decisions evaluates up to 400 bounded yes-or-no, multiple-choice or scored rubric questions per request up to 10 times faster than the Responses API, with zero output token charges.
Spotted an error? Email aiforsuccess.in+compass@gmail.com or tag @ai_for_success on X