Jiantao Jiao says an NVIDIA team built SWE-Serve to test whether AI agents can develop an inference engine and serve real models. The benchmark has 53 tasks drawn from real sgl_project engineering work.
Wispr Flow users can now write Hinglish in English letters, in the same form they would use when typing.
The Tev1-4B-experimental weights are in the togethercomputer repository on Hugging Face. Together AI describes the model as an experiment meant to show how specialized decision models can be trained fairly easily.
The Jev-like classifier is finetuned on Qwen3.5 4B and priced at $0.042/M input and $0/M output. Together AI is also releasing a data recipe and a finetuning tutorial, and says training Tev1 cost $17.
OpenRouter now hosts Qwen3.8 Max Prime from Alibaba_Qwen, describing it as a higher-throughput variant of Qwen3.8 Max with the same 2.4T-parameter capabilities.
On September 23, 2026, CoreWeave said SemiAnalysis_ had published ClusterMAX 3.0 that day and that CoreWeave is the only provider rated Platinum in all three editions.
Existing subscribers’ one-time Claude Code cloud-session credit ($100 on Pro or $250 on Max) applies automatically and is separate from usage limits, with one credit per account for cloud sessions only. After a local limit is reached, cloud work can continue until the credit runs out.
ClaudeDevs says cloud sessions are officially available and can keep Claude Code working while a laptop is closed. Existing subscribers get a one-time credit of $100 on Pro or $250 on Max.
On OpenCode v2, OpenChamber 2.0 makes saved skills, agents, MCP servers, and plugins take effect immediately. The release also adds Code Mode, web search cards, and a stats page.
Token Meter is a free, open-source menu bar tool from Splunk that shows coding agent costs as a run is underway.
Agentic workflows can run entirely on a local machine through Gemma 4 and LiteRT. The SDK also supports OpenAI-compatible endpoints, including Ollama, llama.cpp, and vLLM.
Polymarket says the tools are being rolled out to physicians in roughly 100 countries, including Uganda, Sudan, Haiti, and Mongolia.
On September 23, 2026, Cursor made both available on Teams and Enterprise plans and said it would include Rollouts usage credits for the next 10 days so users can try Rollouts on real changes.
Cursor says the update reduces Security Reviewer's average time to 3.8 minutes from 4.8 minutes.
Rollouts write a monitoring plan and watch changes as they deploy, verifying deployments so regressions are caught before users see them.
GitHub says the view remains fast and smooth for a pull request with 2,200 files, more than 1 million changed lines, and more than 400 inline comments.
WebCraftBench has agents use a live app, then scores aesthetics, usability, and whether the original request was met. Tencent Hy reports an 85.3% match with human preference on 197 validated pairs.
Cline says the stealth model is free in Cline, with multimodal input and a 1M-token context window. It also describes the model as having fast inference and strong coding capabilities.
A Codex planner and creator turn each skill into tested terminal workflows, using a rubric taken from that skill's own quality criteria. Outcome-only RL moved Qwen3.8-27B from 49.4% to 54.1% on Terminal-Bench 2.1, while adding the rubric reached 50.1% but won more trajectory comparisons.
Arena.ai says it has generated side-by-side outputs from Anthropic’s Claude Opus 5.5 and OpenAI’s GPT-6 Sol, and that scores for Claude Opus 5.5 are coming soon.
