https://tpsreport.news/news/google-elephant-goldfish-model-plugin-spec-first-coding TPS: Tokens Per Second en 2026-11-08T11:35:34.996Z Google's Elephant-Goldfish Model plugin has coding agents draft specs, then run 5 zero-context review stages before code Google, coding agents, spec-first coding, plugin, Elephant-Goldfish Model, developer tools https://tpsreport.news/news/moe-routing-comparison-deepseek-qwen-mixtral-llama-4-gpt-oss TPS: Tokens Per Second en 2026-11-08T10:35:55.542Z MarkTechPost Compares Token Routing Across 10 Open MoE Models, Including DeepSeek, Qwen, Mixtral, Llama 4 and gpt-oss mixture-of-experts, MoE, routing, DeepSeek, Qwen, Mixtral, Llama 4, gpt-oss https://tpsreport.news/news/co-design-bench-coding-agents-26-percent-fully-correct TPS: Tokens Per Second en 2026-11-08T10:05:47.088Z Co-Design Bench: best coding agent fully solves only 26.4% of real user change requests Co-Design Bench, coding agents, benchmark, GPT-6 Astra, Claude Opus 5, Qwen3.8-Max, Alibaba, agent harness https://tpsreport.news/news/tencent-workrally-rally-video-edit-upscale-models TPS: Tokens Per Second en 2026-11-08T02:06:08.311Z Tencent adds three video models to WorkRally: Rally-Video, Rally-Video-edit-2.0 and UpScale-rally with 4K output Tencent, WorkRally, Rally-Video, video generation, video editing, super-resolution, UpScale-rally https://tpsreport.news/news/huawei-ascend-kimi-k3-1-pro-lora-fine-tuning-atlas-950-superpod TPS: Tokens Per Second en 2026-11-08T02:05:45.831Z Huawei Ascend validates LoRA fine-tuning of 2.8T-parameter Kimi K3.1 Pro on 128-card Atlas 950 SuperPoD Huawei, Ascend, Kimi K3.1 Pro, Moonshot AI, LoRA, fine-tuning, Atlas 950 SuperPoD, mixture-of-experts https://tpsreport.news/news/claude-personal-finance-vs-chatgpt-finances-privacy TPS: Tokens Per Second en 2026-11-07T23:20:47.977Z Claude Personal Finance skips model training by default; ChatGPT Finances follows your opt-out setting Claude, Anthropic, ChatGPT, OpenAI, Personal Finance, Privacy, Plaid, Intuit https://tpsreport.news/news/openai-disproves-exponential-time-hypothesis TPS: Tokens Per Second en 2026-11-07T17:35:38.297Z OpenAI disproves Exponential Time Hypothesis, co-proposer Impagliazzo calls it "shocking news" OpenAI, Exponential Time Hypothesis, computational complexity, research, encryption, chip design https://tpsreport.news/news/inkling-2-small-profanity-vals-ai-benchmark-logs TPS: Tokens Per Second en 2026-11-07T16:50:43.593Z Vals AI finds 316 expletives in Thinking Machines' Inkling 2 Small reasoning logs across 2,655 benchmark runs Thinking Machines, Inkling 2 Small, Vals AI, benchmark, chain of thought, ProgramBench https://tpsreport.news/news/vals-cyberoffense-gpt-6-sol-daybreak-deepseek-v4-1-flash TPS: Tokens Per Second en 2026-11-07T15:50:50.088Z Vals AI: OpenAI's GPT-6 Sol Daybreak hits 80.8% on CyberOffense, open-weight Deepseek V4.1 Flash reaches 72.8% Vals AI, CyberOffense, GPT-6 Sol, Deepseek V4.1 Flash, OpenAI Daybreak, open-weight models, cybersecurity, AISI https://tpsreport.news/news/nous-research-arete-27b-qwen3-8-propaganda-fine-tune TPS: Tokens Per Second en 2026-11-07T12:50:53.702Z Nous Research releases Arete-27B, a Qwen3.8-27B fine-tune that cuts propaganda rate from 22% to 0.9% Arete-27B, Nous Research, Qwen3.8-27B, fine-tuning, open-weight, AI bias, Alibaba, Hugging Face https://tpsreport.news/news/into-the-unknown-benchmark-ai-agents-national-labs TPS: Tokens Per Second en 2026-11-07T11:20:52.474Z None of 135 AI agent submissions pass experts' bar on open national-lab research in 'Into the Unknown' benchmark, AI agents, autonomous research, Into the Unknown, Carnegie Mellon, GPT-6 Astra, Claude Opus 5.5, GPT-5.6 Sol https://tpsreport.news/news/google-research-smartwatch-hypertension-detection-no-cuff TPS: Tokens Per Second en 2026-11-07T10:05:42.615Z Google Research AI flags high blood pressure from smartwatch pulse and motion data, no cuff calibration needed Google Research, Google Health, hypertension, smartwatch, SensorFM, Nature, health AI, wearables https://tpsreport.news/news/prunaai-p-h3-image-8mp-8-steps-minimax-h3 TPS: Tokens Per Second en 2026-11-07T08:50:54.004Z PrunaAI releases P-H3-Image: 8 MP image generation and editing in 8 steps, built on MiniMax-H3 P-H3-Image, PrunaAI, MiniMax-H3, image generation, image editing, diffusers, Hugging Face, ComfyUI https://tpsreport.news/news/opencsg-agentic-27b-tool-workflows-single-a800 TPS: Tokens Per Second en 2026-11-07T07:21:14.143Z OpenCSG Agentic-27B: 27.36B-Parameter Tool-Use Model Runs on One A800, Commercial Use Needs Permission OpenCSG, Agentic-27B, Qwen3.8-27B, tool use, agents, LoRA DPO, open weights, NVIDIA A800 https://tpsreport.news/news/huawei-harmonyos-pc-standalone-assistant-work-office-agent TPS: Tokens Per Second en 2026-11-07T07:20:48.194Z Huawei releases standalone HarmonyOS PC assistant (v11.7.9.300) with official Work office agent Huawei, HarmonyOS, AI agent, office automation, HarmonyOS PC, Work agent, AppGallery https://tpsreport.news/news/meta-muse-active-memory-active-focus-specialized-models TPS: Tokens Per Second en 2026-11-07T02:05:39.033Z Meta brings Active Memory and Active Focus to all Muse and Meta AI users Meta, Muse, Meta AI, Active Memory, Active Focus, Muse Wonder https://tpsreport.news/news/github-in-slack-and-teams-generally-available TPS: Tokens Per Second en 2026-11-07T00:05:41.612Z @GitHub in Slack and Teams is now generally available GitHub, GitHub Copilot, Slack, Microsoft Teams, product update, pull requests https://tpsreport.news/news/flux-3-image-klein-nvidia-b200-tensorrt-llm-visualgen TPS: Tokens Per Second en 2026-11-06T22:36:52.003Z NVIDIA: FLUX 3 Image Klein Base runs up to 3.46x faster on one B200 with TensorRT LLM VisualGen FLUX 3, Black Forest Labs, NVIDIA, Blackwell, B200, TensorRT LLM, NVFP4, image generation https://tpsreport.news/news/cursor-voice-chat-agents-grok-voice TPS: Tokens Per Second en 2026-11-06T22:36:25.647Z Cursor adds realtime voice calls with coding agents, powered by Grok Voice Cursor, Anysphere, voice chat, Grok Voice, AI coding agents, iOS, Privacy Mode https://tpsreport.news/news/raycast-dictate-to-ai-screen-awareness-v2-11 TPS: Tokens Per Second en 2026-11-06T22:35:59.450Z Raycast adds Dictate to AI, a voice command that sends active-app context to AI Chat Raycast, Dictate to AI, Screen Awareness, AI Chat, voice dictation, macOS, product update https://tpsreport.news/news/anthropic-project-eve-claude-trading-sabotage-detection TPS: Tokens Per Second en 2026-11-06T21:05:38.441Z Anthropic's Project EVE: Claude agents profit in simulated EVE Online markets but rarely catch a sabotaging teammate Anthropic, Claude, Project EVE, EVE Online, EVE Frontier, multi-agent, sabotage detection, Anthropic Fellows Program https://tpsreport.news/news/openai-biobasicsbench-wet-lab-agent-benchmark TPS: Tokens Per Second en 2026-11-06T19:20:39.809Z OpenAI introduces BioBasicsBench, a 100-task benchmark for AI agents in everyday wet-lab science OpenAI, BioBasicsBench, benchmark, AI agents, wet-lab, biology https://tpsreport.news/news/anthropic-model-4-aeci-178-metr-report TPS: Tokens Per Second en 2026-11-06T18:35:55.392Z METR report: Anthropic's unreleased Model 4 scores 178.6 on AECI, ahead of Opus 5.5 Anthropic, METR, AECI, Opus 5.5, Model 4, benchmark, AI R&D, Responsible Scaling Policy https://tpsreport.news/news/meta-muse-spaces-custom-app-experiences TPS: Tokens Per Second en 2026-11-06T18:20:51.507Z Meta adds "spaces" to Muse AI, letting users build app-like hubs by describing them Meta, Muse, Muse Wonder, AI assistant, spaces, product update https://tpsreport.news/news/salesforce-koa-decisions-gemma-image-audio TPS: Tokens Per Second en 2026-11-06T16:50:48.763Z Salesforce launches Koa Decisions, a Gemma-based model that classifies text, image and audio requests Salesforce, Koa Decisions, HyperClassifier, Gemma, Agentforce, multimodal https://tpsreport.news/news/google-antigravity-2-29-remote-agents-voice-mode TPS: Tokens Per Second en 2026-11-06T15:50:49.819Z Google Antigravity 2.29 adds cloud Remote Agents for Google AI Pro and Ultra subscribers Google, Antigravity, Remote Agents, Voice Mode, Gemini Live, agentic coding, Google AI Pro, Google AI Ultra https://tpsreport.news/news/underdog-woof-31m-parameter-open-text-to-speech-model TPS: Tokens Per Second en 2026-11-06T13:21:02.001Z Underdog releases Woof, a 31M-parameter open text-to-speech model that runs in a browser Underdog, Woof, text-to-speech, on-device AI, open models, edge AI, local AI, Bark