Tags: agents
51 article(s)
MiniMax launches M3.1 Flash at 0.15 dollars per million input tokens, FLUX 3 Max generates video 22 times faster than Seedance 2.5, Meta details Prometheus.
Anthropic commits 5 % of its shares to the public good, OpenAI stops billing for idle environments, Thomas Wolf puts a figure on compute for open models.
Anthropic studies its jobs and employee expectations, DeepSeek releases V4.1-Flash-1017 and its Search and Fetch APIs, Claude Code responds faster.
Claude explains the seeds for NIST ECDSA curves, Beam gets open weights, Gemini 4 Argon beats Minecraft and GPT-6 Astra Pro goes multi-agent.
Muse Spark Open, an open MoE with 816 billion parameters from Meta, Nemotron 3.5 Super VL from NVIDIA, Cohere to acquire AISLE, and SPACE goes open source.
Grok Voice 3.0 leads at $0.08 per minute, Gemini 4 Argon in science and robotics, Claude Code with parallel sessions, ten medical manuscripts from Meta.
Gemini 4 Argon available to everyone at 2 dollars per million tokens, The X Subscription from 0 to 200 dollars per month, Run:ai as open source and LFM2.5-24B-A2B.
Muse Wonder from Meta at 4 dollars per million tokens, Vera Rubin rack designs contributed to Open Compute, replication by small models studied by Anthropic.
MASS from Sakana AI and UC Berkeley: 69,9 % wins against the starting model on held-out tasks, a quantum competition, Qwen Code v0.25.1.
Unintended actions by Claude on real websites, Internet access cut for all internal evaluations, Amp opens up to Claude subscriptions, vLLM on Vera Rubin.
Google Cloud Gemini agent, HeyGen Voice tops the Controlled Voice ranking, 1 000 agents per run in Claude, Codex predicts the next message.
Claude Dashboards and Motion in beta, Docs and Slides for everyone, Cyber Mission and OSS Scanner, GPT-6.1 Sol with Ultrafast, Genesis Mission and Nano Banana 2.1.
GPT-6 and Intelligent UI for everyone in ChatGPT, Claude Haiku 5.5 at 0.10 dollar, RTX Spark PC preorders, local Copilot, 722 mathematical manuscripts.
Mistral Large 4 in preview, Claude in Google Docs, Sheets and Slides, a three-tier cyber program, EmbeddingGemma 2 and cheaper decision models.
Invisible watermark for ChatGPT and Codex text in the European Union, memory across sessions for Devin, GitHub ReviewBench, Cohere North 2 and Beam.
ThinkingBox replays 507 tasks 20 times on OpenEnv and agent rankings shift, 10,000 Exgentic traces in OpenTelemetry format, Claude Code 2.1.289.
Kolibri, a 78-billion-parameter English-German model under Apache 2.0, Meta safety rules extended to training, Devin in Jira.
Six mathematics papers written with Muse Spark, a Perplexity API that answers with probabilities, 100 million dollars to train 10,000 engineers.
Mods customize Claude Code in TypeScript, GitHub Copilot controls desktop apps and orchestrates its agents, FLUX 3 Image generates in 4K.
Gemini 4 Argon arrives for cyber defenders, OpenAI says it thwarted a distillation campaign, Cohere launches Embed 5, Ideogram 4.5 and HeyGen Video.
DevDay 2026: OpenAI launches GPT-6.1 Sol, dots and Codex cloud; Clément Delangue announces NVIDIA’s acquisition of Hugging Face, AMD to acquire World Labs.
Anthropic launches Claude Sonnet 5.5, faster and cheaper per task than Sonnet 5; NVIDIA puts safeguards around agents down to the silicon, and Manus moves to 2.0.
September 22: Anthropic launches Claude Opus 5.5 and moves Pro and Team Standard to Opus in Claude Code, OpenAI releases GPT-6 Sol and GPT-6 Luna at half the promotional price of GPT-5.6, and both families arrive the same day in GitHub Copilot, Devin, and Perplexity.
September 21: xAI releases Grok 4.7 at the same price as Grok 4.6 and sees it deployed the same day in Cursor and GitHub Copilot, OpenAI reveals an internal model that solved more than 100 open mathematics problems, and Google opens Googlebook preorders starting at $899.