https://tpsreport.news/news/ai2-fairstress-llm-fairness-benchmark TPS — Tokens Per Second en 2026-10-20T16:20:51.940Z Ai2 releases FairStress: 13M-case benchmark finds LLM fairness gaps hidden by final answers FairStress, Ai2, Olmo-eval, fairness, benchmark, bias, LLM evaluation, adversarial prefixes https://tpsreport.news/news/mistral-voxtral-3-speech-models-voice-agent-sdk TPS — Tokens Per Second en 2026-10-20T16:06:00.291Z Mistral releases Voxtral 3: 50-language transcription, expressive TTS and a real-time voice agent SDK Mistral AI, Voxtral 3, speech-to-text, text-to-speech, voice agents, diarization, multilingual, EU sovereignty https://tpsreport.news/news/perplexity-computer-local-mac-36gb-qwen-3-6-pplx-pro TPS — Tokens Per Second en 2026-10-20T15:50:50.330Z Perplexity Computer now runs fully on-device on Macs with 36GB+ RAM, using Qwen 3.6 or PPLX Pro Perplexity, Perplexity Computer, local AI, on-device AI, Qwen 3.6, PPLX Pro, Apple Silicon, macOS https://tpsreport.news/news/ibm-altk-evolve-gpt-4-1-tau2-bench-retail-67-5-pass-4 TPS — Tokens Per Second en 2026-10-20T15:35:47.272Z IBM's ALTK-Evolve lifts GPT-4.1 to 67.5% Pass^4 on τ²-bench Retail without weight updates ALTK-Evolve, IBM Research, τ²-bench, GPT-4.1, AI agents, customer service, benchmark, agent memory https://tpsreport.news/news/cloudflare-clairic-document-extraction-workers-ai-preview TPS — Tokens Per Second en 2026-10-20T12:21:13.088Z Cloudflare releases Clairic, a single-pass document extraction model, in preview on Workers AI Cloudflare, Clairic, Workers AI, document extraction, open weights, agents, structured data https://tpsreport.news/news/openai-cobol-migration-multilangbench-gpt-6-astra TPS — Tokens Per Second en 2026-10-20T12:20:50.341Z OpenAI claims GPT-6 Astra passes 87.4% of MultiLangBench tests, 72% on COBOL OpenAI, COBOL, legacy code, code migration, MultiLangBench, GPT-6, benchmarks, enterprise https://tpsreport.news/news/amazon-bedrock-claude-structured-outputs-strict-tool-use TPS — Tokens Per Second en 2026-10-19T23:50:46.770Z Amazon Bedrock adds schema-enforced structured outputs for all Anthropic Claude models on both endpoints amazon-bedrock, anthropic, claude, structured-outputs, tool-use, aws, constrained-decoding, json-schema https://tpsreport.news/news/mercury-3-diffusion-llm-1m-context TPS — Tokens Per Second en 2026-10-19T21:20:56.242Z Inception launches Mercury 3 diffusion LLM: 1M context, 1,068 tokens/sec, $0.20/$0.70 per 1M tokens Mercury 3, Inception, diffusion LLM, low latency, 1M context, OpenRouter, AI agents, model release https://tpsreport.news/news/minimax-m3-1-flash-1m-context-open-weights TPS — Tokens Per Second en 2026-10-19T18:35:53.919Z MiniMax M3.1 Flash scores 42 on AA Intelligence Index at $0.15/$0.60 per 1M tokens, open weights Oct 23 MiniMax, M3.1 Flash, open weights, reasoning model, Artificial Analysis, DeepSeek, pricing, China AI https://tpsreport.news/news/kling-4-0-30-second-video-stereo-audio-flash-variant TPS — Tokens Per Second en 2026-10-19T17:50:55.132Z Kling AI releases 4.0: 30-second video, stereo audio, 10+ language lip sync, plus a 720p Flash tier Kling AI, Kling 4.0, Kling 4.0 Flash, video generation, stereo audio, lip sync, MCP, Wan 3.0 https://tpsreport.news/news/timestral-209m-time-series-forecasting TPS — Tokens Per Second en 2026-10-19T17:20:57.665Z Mistral releases Timestral-209M, a time series model claiming top zero-shot scores on three benchmarks Mistral AI, Timestral, time series, forecasting, foundation model, open weights, GIFT-Eval, fev-bench https://tpsreport.news/news/nvidia-nemotron-3-5-super-vl-120b-a12b-nvfp4 TPS — Tokens Per Second en 2026-10-19T15:36:02.637Z NVIDIA releases Nemotron 3.5 Super VL NVFP4: 120B-parameter, 12B-active vision-language model with 1M context NVIDIA, Nemotron, vision-language, NVFP4, mixture-of-experts, Mamba-2, open-weights, quantization