The New Stack

Signals

Latest developments

September 2026

◆ Anthropic rolled out a unified Claude interface that merges chat with its Cowork agent and added Docs and Slides capabilities Feature update

◆ Google releases Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, new near-real-time speech-to-speech voice dialogue models Model release

◆ New coding and agent benchmarks using private, tool-enabled tasks show large drops in pass rates and limited consistency across repeated runs Benchmark result

◆ Perplexity adds a local “Portable Computer” agent to its Windows app for compatible NVIDIA RTX PCs with 24GB+ VRAM Deployment

◆ Amazon Bedrock prompt caching capability for reusing partially processed prompt context to reduce billed input tokens and improve time-to-first-token Feature update

◆ Cohere released North Small Translate, an open-weight mixture-of-experts machine translation model available via API and with non-commercial self-hosting checkpoints Model release

◆ Salesforce launched its Enterprise AI Harness and AI Control Plane to help organizations build and manage AI agents across its platform Product launch

◆ Security researchers report that a significant share of Model Context Protocol (MCP) access policies are broken or missing Security issue

◆ Agents on Rails benchmark leaderboard updated with rerun results for Anthropic Claude Fable 5.1 and Z.ai GLM 5.3 Flash Benchmark result

◆ New benchmarks and benchmark results were published to evaluate AI agents’ end-to-end commercial performance and their ability to build other agents Benchmark result

◆ Anthropic researcher Jacob Coxon resigns and warns that frontier AI could pose existential risk Executive move

◆ OpenAI rolled out an automated security review that can block merges of pull requests from its engineers Feature update

◆ Mistral AI raises a €3 billion Series D round led by Samsung Electronics at a valuation above €21 billion Funding

◆ MBZUAI IFM releases K2 Horizon, a suite of six Apache 2.0 open foundation models with training artifacts and deployment support Open source release

◆ OpenAI and Anthropic researchers publicly warned that current AI progress may outpace alignment controls and urged stronger safeguards, including calls for slowing development Other

◆ OpenAI launched the Agents API in public beta and reported increased internal use of coding “research intern” agents for accelerating research experiments Product launch

◆ OpenAI launches GPT-6 Astra and begins a staged rollout to selected users and partners, followed by broader availability Model release

◆ OpenAI released GPT-6 Astra, its flagship AI model, along with evaluation results, a system card, and related updates to how the model is benchmarked and priced Model release

• 16 Sep Perplexity’s AI agents helped build a database. They weren’t allowed to run it.
• 16 Sep “Everyone’s in a race to replace GitHub”: Zed launches Delta because agents made pull requests obsolete
• 16 Sep Anthropic bet users were choosing wrong. So it removed the choice.
• 15 Sep OpenAI’s voice model doesn’t think. That’s the point.
• 15 Sep AWS agents will suggest your new flights. Code decides what gets booked.
• 15 Sep Bolt is giving developers 50x more compute. But there’s a catch.
• 15 Sep Agents operate, humans govern: Scale your operations and reduce toil with Azure SRE Agent
• 14 Sep AI’s best coding agent fails 60% of the time — and the data backs it up
• 14 Sep Perplexity’s new agent runs entirely on your GPU — with one expensive catch
• 14 Sep Your AI coding spend bought 25% more output. Duplication rose 81%.
• 14 Sep Chinese AI models dominate OpenRouter’s US token consumption. It can now guarantee that traffic stays entirely in the US.
• 14 Sep Why an old caching trick is your secret to lower LLM costs
• 13 Sep Chip Huyen explains how to cut inference costs without new hardware
• 13 Sep “Machine translation is still broken for most of the world’s languages”: Cohere builds non-reasoning for a reason
• 13 Sep It passed CI. It passed your evals. The customer still got the wrong answer.
• 12 Sep Why MCP security is about permissions overhaul
• 12 Sep The AI-native SDLC won’t be one process
• 12 Sep Jacob Coxon warns AI could kill us all. Anthropic’s own report exposes safety gaps.
• 11 Sep OpenAI’s researchers burned $7,000 a day on AI agents — now it’s opening the floodgates
• 11 Sep OpenAI’s safety system is already cutting off API responses mid-task
• 11 Sep Cohere’s new translation model is open weights — but not for commercial use
• 11 Sep Kubernetes v1.37 brings 67 enhancements. Which matter for operators?
• 11 Sep How AWS Lambda logs every flow across thousands of microVMs per host with eBPF and Rust
• 10 Sep “Six tools, one harness”: Salesforce loops together a six-pack of favorites
• 10 Sep “Valuable warning shots”: How Anthropic now views Claude’s cyber incidents
• 10 Sep Mistral wants open-weight AI to compete at the frontier. It just raised $3.5 billion to do it.
• 10 Sep Researchers found that 1 in 5 MCP access policies came back broken or missing
• 10 Sep Fable 5.1 vs. Fable 5: Results on a real-world budget, not the spec sheet
• 10 Sep 47,000 job listings reveal the engineering roles that AI is creating
• 10 Sep Stop AI code sprawl before it destroys your software design
• 10 Sep AI floods security teams with flaws — business context sets priorities
• 10 Sep Nvidia and Palantir fine-tune a 30B Nemotron model for Nvidia’s supply chain. It beats a model 18 times its size.
• 9 Sep Claude did best on a new benchmark for ‘agents that build agents’. It still passed fewer than a quarter of the tests.
• 9 Sep How much control should AI get? A CISO roundtable takes on SOC autonomy
• 9 Sep K2 Horizon just shipped as six new fully open models — developers aren’t fully convinced
• 9 Sep Harness rebuilt its Git repository for nonstop AI agent traffic
• 8 Sep DeepSeek is hiring 150 engineers, and none of them will touch a model
• 8 Sep AI broke code review. Two experts disagree on what replaces it.
• 8 Sep After nine years as HashiCorp CEO, Dave McJannet now wants to “unblock” enterprise AI agents
• 7 Sep “Some agents will be pursuing their own objectives”: OpenAI’s chief scientist warns AI could trick and blackmail humans
• 7 Sep AI agents are creating more work, not less — and OpenAI’s own numbers back it up
• 7 Sep OpenAI’s new model costs 2.5x more per token — and developers are saving money anyway
• 7 Sep “Twenty years of brand building simply froze in time”: How coding agents select their tools of choice
• 7 Sep Permissions belong in the assembly context
• 6 Sep Polars 2.0 pre-release comes with a 5x speed boost — but it could change row order
• 5 Sep Building trust in agentic RAG starts with evidence
• 4 Sep Microsoft built a prompt injection detector. Then it caught a phishing campaign instead.
• 4 Sep “Sorry for the messy rollout”: OpenAI launched GPT-6 Astra, but developers are locked out
• 4 Sep OpenAI will sell you Astra, but not the system that scored 98.6% on ARC-AGI-3
• 4 Sep AI agent evaluations are part of the product

Relationships

Partnerships

• Partnered with Anthropic · 3 sources
• Partnered with GLM-5.3 · 1 source
• Partnered with OpenRouter · 1 source

Products & technology

• Integrated with GLM-5.3-Flash · 1 source
• Integrated with GLM-5.3 · 1 source
• Supplies Junie Local · 1 source
• Derived from Solar Pro 4 · 1 source
• Kimi integrated with this company · 1 source