https://tpsreport.news/news/openai-misalignment-reports-model-hacked-government-site-hospital-data TPS — Tokens Per Second en 2026-10-17T15:50:50.362Z OpenAI reports models forged tokens and pulled 24 years of hospital data from a government site during testing OpenAI, AI safety, misalignment, GPT-5.6-Sol, agent security, reward hacking, chain-of-thought, AI research https://tpsreport.news/news/claude-solves-nsa-ecdsa-curve-seed-mystery TPS — Tokens Per Second en 2026-10-17T14:50:51.805Z Claude traces NSA-supplied NIST curve seeds to lrand48 generator in under 30 minutes, per Anthropic Anthropic, Claude, cryptography, NSA, NIST, ECDSA, security, research https://tpsreport.news/news/ai-models-hide-mistakes-work-reports-honesty-prompt-study TPS — Tokens Per Second en 2026-10-17T10:05:47.601Z Study: GPT-5.5 flags a hidden negative result in 2 of 200 abstracts; one honesty sentence lifts it to 190 research, AI honesty, GPT-5.5, Gemini 3.1 Pro, Claude Opus 4.8, Qwen3.5-9B, AI agents, prompt engineering https://tpsreport.news/news/deepseek-v4-1-flash-1017-552b-ced-1m-context TPS — Tokens Per Second en 2026-10-17T08:35:54.149Z DeepSeek V4.1 Flash 1017 launches with 552B CED backbone, 1M context, $0.15/$0.60 per 1M tokens DeepSeek, DeepSeek V4.1 Flash, mixture-of-experts, 1M context, multimodal, agents, OpenRouter, model release https://tpsreport.news/news/nvidia-qwen3-8-flash-next-fp4-local-nvfp4-checkpoint TPS — Tokens Per Second en 2026-10-17T01:35:50.679Z NVIDIA releases 96 GiB NVFP4 build of Qwen3.8-Flash-Next for single-GPU Blackwell and DGX Spark NVIDIA, Qwen, Alibaba, NVFP4, quantization, vLLM, DGX Spark, Blackwell https://tpsreport.news/news/elorian-glimpse-visual-reasoning-model-api-pricing TPS — Tokens Per Second en 2026-10-16T16:20:49.619Z Elorian launches Glimpse, a visual reasoning model priced at $0.60/$3 per 1M tokens Elorian, Glimpse, multimodal, visual reasoning, video understanding, robotics, research preview, seed funding https://tpsreport.news/news/reflection-beam-501b-open-weight-agentic-coding-aws TPS — Tokens Per Second en 2026-10-16T15:35:54.373Z Reflection AI releases Beam, a 501B-parameter open-weight MoE coding model under Apache 2.0 Reflection AI, Beam, open-weight, mixture-of-experts, agentic coding, AWS, SageMaker JumpStart, Apache 2.0 https://tpsreport.news/news/gemini-4-argon-arc-agi-3-benchmark TPS — Tokens Per Second en 2026-10-16T15:20:49.453Z Gemini 4 Argon scores 97.5% on ARC-AGI-3 with vendor features, but only 49.8% in the standard harness Gemini 4 Argon, Google DeepMind, ARC-AGI-3, Epoch AI, benchmarks, GPT-6 Astra, ARC Prize https://tpsreport.news/news/transformers-v5-20-1-reflection-ai-beam TPS — Tokens Per Second en 2026-10-16T13:20:46.465Z Hugging Face Transformers v5.20.1 adds Reflection AI's Beam model and fixes loading on older torch Hugging Face, Transformers, Reflection AI, Beam, open-source, coding models, agentic AI, release https://tpsreport.news/news/cline-v4-1-25-cloud-sessions-gemini-4-argon-default TPS — Tokens Per Second en 2026-10-16T06:50:45.999Z Cline v4.1.25 adds cloud sandbox sessions and makes Gemini 4 Argon the default model for its own provider Cline, coding agents, Gemini 4 Argon, cloud sessions, release notes, OpenRouter, Ollama, LM Studio https://tpsreport.news/news/goodfire-cammarata-neural-networks-already-clean-polysemanticity TPS — Tokens Per Second en 2026-10-16T04:35:52.237Z Goodfire researcher claims 90% of Qwen-32B neurons have one job, challenging polysemanticity interpretability, alignment, polysemanticity, Goodfire, Qwen, mechanistic-interpretability, agents, research https://tpsreport.news/news/nvidia-nemotron-pretraining-indic-v1-2-8t-tokens TPS — Tokens Per Second en 2026-10-16T01:50:47.591Z NVIDIA and AI4Bharat announce Nemotron-pretraining-Indic-v1: 2.8T tokens across 22 Indian languages NVIDIA, Nemotron, Indic languages, AI4Bharat, synthetic data, pretraining data, NeMo Data Designer, multilingual https://tpsreport.news/news/gemini-4-argon-antigravity-ai-pro-ultra-rollout TPS — Tokens Per Second en 2026-10-16T01:20:47.096Z Google rolls out Gemini 4 Argon to Antigravity for AI Pro and Ultra subscribers with separate weekly quota Gemini 4 Argon, Google, Antigravity, AI Pro, AI Ultra, AI coding, agentic development, Gemini 3.8 Flash https://tpsreport.news/news/chatgpt-chrome-extension-open-tabs-access TPS — Tokens Per Second en 2026-10-15T22:50:45.890Z OpenAI's ChatGPT Chrome extension can now read your open tabs and control them via ChatGPT Work OpenAI, ChatGPT, Chrome extension, ChatGPT Work, browser agents, ChatGPT Atlas, Google Chrome, product update https://tpsreport.news/news/moondream-3-preview-9b-moe-vision-model TPS — Tokens Per Second en 2026-10-15T22:36:18.928Z Moondream 3 Preview lists on OpenRouter: 9B MoE vision model, 2B active, 32K context, $0.30/$2.50 Moondream, Moondream 3 Preview, vision-language model, mixture-of-experts, OpenRouter, OCR, multimodal, Prime Intellect https://tpsreport.news/news/moondream-3-1-9b-moe-vision-language-32k-context TPS — Tokens Per Second en 2026-10-15T22:35:48.951Z Moondream 3.1: 9B MoE vision model with 2B active parameters, 32K context, $0.30/$2.50 per 1M tokens Moondream, Moondream 3.1, vision-language model, mixture-of-experts, OCR, OpenRouter, multimodal, model release https://tpsreport.news/news/ibm-granite-switch-4-1-8b-30b-decision-speech-one-endpoint TPS — Tokens Per Second en 2026-10-15T20:51:01.144Z IBM Granite Switch 4.1 8B and 30B fold chat, decisions and speech-to-text into one endpoint IBM, Granite, Granite Switch, aLoRA, speech-to-text, ASR, guardrails, vLLM https://tpsreport.news/news/anthropic-python-sdk-v1-15-0-speed-parameter TPS — Tokens Per Second en 2026-10-15T19:50:50.045Z Anthropic Python SDK v1.15.0 adds speed parameter to Messages API and usage.speed field Anthropic, Python SDK, Messages API, changelog, SDK release, extended thinking, OpenTelemetry https://tpsreport.news/news/ollama-v0-40-5-nvidia-nemotron-3-5-super-bolt TPS — Tokens Per Second en 2026-10-15T19:05:54.444Z Ollama v0.40.5 adds NVIDIA Nemotron 3.5 Super, a 120B MoE vision-language model with up to 1M-token context ollama, nvidia, nemotron, nemotron-3.5-super, nemotron-3.5-bolt, local-inference, mixture-of-experts, llama.cpp