2026
10,781 articles · 136 days · 10 categories · avg 79 per day
Bank Tech Lead Grows OpenCode Telegram Bot to 1,298 GitHub Stars With AI Agents
A bank tech lead used AI agents to grow an open-source OpenCode bot to 1298 stars and 48 contributors in a year. Here is what worked in development and promotion.
Baseline on Bukva recognizes Russian fingerspelling letters with 83.6% top-1 accuracy
The WSLP workshop at EMNLP 2026 showcased two Russian fingerspelling datasets and the HandReader models. A lightweight baseline on Bukva gets the letter right with…
Plurio Agent Brought Neobank Finom 38% More Customers While Ad Spend Fell 8%
Finom handed management of more than 500 ad creatives to the Plurio AI agent. Over five months, new customer flow rose 38% and ad spend fell 8%.
WeFair-2B Query Rewriter Lifts CTR by 2.951% in WeChat Short-Video Search
TokenRouter speeds up token-level model routing by up to 64x
NVIDIA Nemotron 3.5 Super VL: 120B Parameters, 12B Active, Running on AWS
Google lets third-party AI agents into Android Studio: Claude, Codex and 40+ more
A Swarm of AI Agents Builds the Game Spawn, and Tokens Cost $3 Million a Month
Startup Spawn handed development of its multiplayer game to agents running on Fable: hourly production deploys, 30-second rollbacks, and tokens that cost 30 times…
MIT: AI agents share KV-cache instead of text and run 4x faster
MIT researchers propose SwaT: agents pass each other already-processed context instead of a summary. Answers are 40% better on average, and speed is four times…
Mistral Releases Large 4 Weights: Open 1T-Parameter Model for Coding
Google unveils Gemini 4 Carbon: 88% on vulnerability discovery and first place in the ranking
Anthropic will give companies token reports: customer spending grew 9x in a year
Stanford AI tool verifies part of Signal's code in half a day for under $500, versus five months of expert work
Qwen 3.5 4B gets a RISC-V processor inside its weights, with DeltaNet as memory
Divyansh Agrawal compiled a 32-bit RISC-V core into Qwen 3.5 4B's weights without training. Long-number arithmetic accuracy rose from 48.6% to 100%, matching Qwen…
NVIDIA releases Nemotron 3.5 Transcribe: 40 languages and 100 hours of audio in 3 minutes
NVIDIA's open model transcribes speech in 40 languages, separates speakers, and processes 100 hours of audio on a single RTX PRO 6000 in under three minutes.
Claude in Houdini on Windows: Two Bugs in the Agent Panel and a 31-Node Geode
The author connected Claude to Houdini via an MCP panel, found two Windows bugs, submitted fixes, and built a procedural geode from 31 nodes in 22 minutes.
Cloudflare releases open-source Clef models: 98.5% on BFCL and 38.8 ms
Gemini 4 Carbon tops independent rankings, but only cyber defenders can use it
How OpenAI Kept Azure Cosmos DB Standing Under the Weight of 1 Billion ChatGPT Users
Modal Passes $1B ARR and Hits a $16B Valuation on Demand for Sandboxes
Ai2 shows LLM novelty judges break on a few words in the prompt
Ai2 tested six LLM novelty judges on 299 ideas from ICLR 2026 reviews: rewording a prompt cut GPT-5.4 accuracy from 93.5% to 40.9%, and costly systems lost to a…
OvisOCR3: an 0.8B-parameter model that reads documents better than specialized systems
Researchers at ATH-MaaS merged five separate document-recognition tasks into one compact model — and it beats specialized OCR systems on most metrics.
Knowledge Conflict in RAG: Why You Need Probability of Context Correctness, Not a Rigid Algorithm
Researchers Find AI Agents Confuse Old and New Facts in Memory Even After Correct Retrieval
Skills for AI Agents Aren't Universal: Weak Models Need Their Own Version of the Prompt
AI agent on GPT-5.5 audits 2600 TravelLine autotests in two weeks
Liquid AI releases LFM3: six multimodal 2B models with a 256K context window
LFM3-1.7B beats LFM2.5-2.6B with 35% fewer parameters and fits in 1 GB at 4-bit. One base model spawns text, vision, audio, omni and calibrated-answer variants.
Black Forest Labs releases FLUX.3 Klein: 1024×1024 in 2 seconds on an H100
FLUX.3 Image Klein from Black Forest Labs is in Diffusers from day one: text-to-image, reference-based editing and bounding-box layout control. One frame takes 1.93…
Google DeepMind finds softmax flaw: gradient drops to zero at a logit gap of 16.6
Barron, Gnanapragasam and Tschannen show that in float32 the cross-entropy gradient vanishes at a gap of about 16.6, and distillation gets spurious noise.
Tencent releases Youtu-Parsing-Omni: one model parses documents, audio and video
Glance: a single 9B model picks its own RAG passages and beats a reranker
Pretrained AI Image Generators Learn to Fill In Any Region at a Fraction of the Cost
