AI News
Daily AI industry briefings - model releases, research, infrastructure, and pricing - summarized neutrally from vetted sources, with every claim linked to its origin.
- Latest
MiniMax releases M3.1 Flash 428B MoE at $0.15/M tokens; NVIDIA open-sources Run:ai GPU orchestration
MiniMax shipped a 428B-parameter mixture-of-experts agentic model with competitive token pricing and 1M context window; NVIDIA released Run:ai GPU orchestration as open source.
MarkTechPostRead → -
Google open-sources Harness2 agent framework; Voxel51 adds MCP tooling for multimodal data
Google released Harness2, a test-time agent refinement framework that improved GLM-5 by 11.5 points without weight updates; Voxel51 3.0 unified multimodal AI data platform with 45+ MCP tools.
MarkTechPostRead → -
DeepSeek V4.1-Flash-1017 cuts agentic costs; StepFun releases 600B MoE Step 5 Preview
DeepSeek released V4.1-Flash-1017 with improved agentic benchmarks and lower per-task costs; StepFun shipped Step 5 Preview open-weight MoE with 1M context window.
The Decoder MarkTechPostRead → -
Meta's Muse Spark Open; NVIDIA Nemotron 3.5 Super VL; Black Forest Labs FLUX 3 Image Klein
Meta released an 816B open-weight MoE model optimized for Apple silicon; NVIDIA and Black Forest Labs shipped enterprise agent and image generation models for on-device deployment.
MarkTechPost The DecoderRead → -
Fish Audio releases 35B open-weight MoE text-to-speech model; TypeSafe AI expands Jev with images and 262k context
Fish Audio released Drama 3, an open-weight mixture-of-experts text-to-speech model with 3B active parameters covering 83 languages; TypeSafe AI expanded Jev 1.14 with image input and extended.
MarkTechPostRead → -
Liquid AI releases 2B-active MoE for on-device agents; Google ships Gemini 4 Argon with 1M token output
Liquid AI released LFM2.5-24B-A2B, an open-weight MoE model optimized for local inference; Google made Gemini 4 Argon generally available with 1M output token capacity.
MarkTechPost Hugging Face Blog Google DeepMindRead → -
Microsoft scales Maia chips to gigawatt inference; agent context-clearing boosts solution quality
Microsoft plans to deploy its in-house Maia 300 chips at multi-gigawatt scale by 2027 for inference; meanwhile, researchers find context-clearing improves autonomous agent performance by up to 66.
The Decoder The VergeRead → -
Token economics shift as demand outpaces cost reductions; Microsoft Decision-1 joins agent routing landscape
Falling AI token prices are driving faster demand growth than cost savings, while Microsoft releases a 9B decision model for agent control and routing.
The Decoder MarkTechPostRead → -
OpenAI Decisions API enters public beta with 10x faster inference; Anthropic orchestrates 1,000 parallel agents
OpenAI's Decisions API launched in public beta with 10x faster typed outputs and new pricing; Anthropic demonstrated multi-agent orchestration catching 94% of bugs versus 39% for single agents.
MarkTechPost The DecoderRead → -
JetBrains Releases Mellum 2.1 Open Coding Agent; Google Brings Agentic AI to Gemini
JetBrains released Mellum 2.1, a 12B open-weight MoE model for coding agents; Google launched agentic capabilities for Gemini targeting enterprise workflows.
MarkTechPost TechCrunch The DecoderRead → -
Claude Haiku 5.5 slashes token prices by up to 90%; AWS Bedrock agent hijacking flaw surfaces
Anthropic cuts Claude Haiku pricing dramatically while researchers expose credential theft vulnerability affecting all AWS Bedrock agents in a region.
The Decoder MarkTechPostRead → -
Mistral Large 4 previews 1T-parameter open-weight MoE; Reflection Beam ships 501B for coding
Mistral AI opened API access to Large 4, a 1T-parameter open-weight MoE with 49B active parameters and 1M context; open weights ship October 31.
AI News RuntimeWire Hugging Face Unite.aiRead → -
llama.cpp v0.6.0 ships 3x Metal speedup; open-weight decision models target agent routing
llama.cpp v0.6.0 delivers 3x Apple Metal throughput with a new batch API; Cloudflare and Amazon ship open-weight decision models for agent routing.
llama.cpp GitHub AI Weekly vLLM GitHub MarkTechPostRead → -
Aleph Alpha releases Kolibri open-weight MoE; Google tightens federated learning privacy
Aleph Alpha released Kolibri, a 78B-parameter open-weight German-English mixture-of-experts model with 3.46B active parameters per token. Google moved federated learning gradient computation into.
MarkTechPost The DecoderRead → -
Aleph Alpha releases Kolibri open-weight MoE; Google tightens Gemini access
Aleph Alpha released Kolibri, a 78.1B-parameter open-weight MoE model with 3.46B active parameters; Google restricts free Gemini access to its weakest tier.
MarkTechPost The DecoderRead → -
Cloudflare releases Clef decision model; Claude Code adds Mods middleware for developer customization
Cloudflare released Clef, an open-weight decision model 10x faster than competitors, while Anthropic added extensibility middleware to Claude Code.
The Decoder TechCrunchRead → -
Open-weight decision models proliferate; Cohere releases Embed 5 for enterprise RAG
Multiple vendors released open-weight decision models optimized for sub-200ms latency, while Cohere launched a new embedding model family targeting enterprise search and agentic retrieval.
MarkTechPost Hugging Face BlogRead → -
OpenAI's GPT-6.1 Sol cuts agentic coding costs to one-fifth of Astra; Perplexity releases contextual embedding model for RAG
OpenAI released GPT-6.1 Sol with near-Astra performance on coding and computer use at significantly lower token costs; Perplexity released a contextual embedding model for RAG pipelines.
MarkTechPost The DecoderRead → -
Perplexity's Photon cuts retrieval latency 92%; Liquid AI releases zero-token decision model
Perplexity shipped Photon, a Rust-based retrieval engine reducing p99 latency from 800ms to 65ms; Liquid AI released d1, a decision model returning calibrated probabilities without generated tokens.
MarkTechPostRead → -
Google Open-Sources RRSI Agent Framework; H Company Releases Holo4 Computer-Use Models
Google released RRSI, enabling LLM agents to self-improve prompts and tools without retraining, while H Company shipped open-weight Holo4 models for desktop, web, and mobile agent automation.
MarkTechPost Hugging Face BlogRead → -
Nvidia embeds agent containment in hardware; Fireworks AI cuts token usage 40% with post-trained Kimi K3
Nvidia launches Open Agent Safety Platform with hardware-level agent isolation; Fireworks AI releases Ember-1 model reducing token consumption by 40% in production.
The Decoder MarkTechPostRead → -
Nvidia releases 100M-parameter speaker diarization model; Stanford team deploys GPT-6 Astra as robot controller
Nvidia released a lightweight open diarization model for real-time speaker identification, while researchers demonstrated direct language model control of humanoid robots without intermediate.
The Decoder The VergeRead → -
Nvidia's SoL-Pi cuts coding agent token usage by 49%; Liquid AI releases speculative decoding for vision-language models
Nvidia's SoL-Pi system reduces coding agent token consumption by up to 49 percent through control layer optimization; Liquid AI ships speculative decoding draft model delivering up to 3.13x faster.
The Decoder MarkTechPostRead → -
Meta's Muse offers cloud Ubuntu environment to 500K users; Black Forest Labs releases FLUX 3 Action robotics model
Meta expands Muse agent with full Ubuntu Linux cloud access for half a million users; robotics model achieves efficiency gains with 7B parameters.
The Decoder TechCrunchRead → -
Contrastive-LM releases CLM-8B agent scoring model; Google ships Gemini 3.8 Flash TTS
Contrastive-LM released CLM-8B, an open-weight agent action scoring model running 9× faster than alternatives; Google released two TTS models supporting prompt-based voice design across 100+.
MarkTechPost Google DeepMindRead → -
Anthropic launches Claude Opus 5.5 at 40% lower cost; OpenAI releases GPT-6 Sol and Luna with halved pricing
Anthropic and OpenAI both released cost-optimized model variants, with Opus 5.5 matching Fable 5.1 performance at significantly lower token prices and Sol/Luna cutting token costs in half versus.
The DecoderRead → -
Hugging Face integrates llama.cpp quantization; Xiaomi's MiMo-V2.6-Pro tops open model rankings
Hugging Face Transformers now supports llama.cpp quantized models for efficient inference; Xiaomi releases MiMo-V2.6-Pro, claiming top performance among open models at lower cost.
Hugging Face Blog The DecoderRead → -
StepFun releases Step 5 Preview 600B MoE model; Alibaba open-sources Qwen-Image-2.1 with 7B parameters
StepFun launched Step 5 Preview, a sparse MoE model with 1M-token context for agentic work; Alibaba released open-weight Qwen-Image-2.1 for image generation on consumer GPUs.
MarkTechPost The Decoder Hugging Face BlogRead → -
Qwen3.8-Omni-Flash matches Gemini multimodal benchmarks at lower cost; Microsoft StudentSim accelerates AI tutor training
Alibaba's Qwen3.8-Omni-Flash delivers multimodal agent capabilities at reduced API pricing, while Microsoft's StudentSim synthetic student framework outperforms GPT-5.4 on tutor training tasks.
The DecoderRead → -
Google Deepmind's Dream-RSI cuts agent iteration by up to 2.43x; Claude used to breach OpenAI systems in under 72 hours
Google Deepmind releases Dream-RSI for efficient agent improvement; security researchers demonstrate Claude exploiting OpenAI vulnerabilities.
The DecoderRead →
Summarized from vetted sources, every claim linked. For information only.