OpenAI's GPT-6.1 Terra Targets Visual Reasoning and Computer Use
GPT-6.1 Terra is priced at $1/million input tokens and $5/million output, almost half the price of GPT-5.6 Terra and GPT-6.1 Sol, and around 10x the price of GPT-6 Luna.
OpenAI emphasizes visual reasoning and computer use. Ted Sanders says Terra is better at technical communication, similar on vision and computer use but much faster, and a bit worse on most other things.
Source: Simon Willison
GPT-6.1 Terra is priced at $1/million input tokens and $5/million output, almost half the price of GPT-5.6 Terra and GPT-6.1 Sol, and around 10x the price of GPT-6 Luna.
OpenAI emphasizes visual reasoning and computer use. Ted Sanders says Terra is better at technical communication, similar on vision and computer use but much faster, and a bit worse on most other things.
Source: Simon Willison
❤2🤮1
Entire launches Git forge with Trails, an agentic replacement for pull requests
Entire, founded by former GitHub CEO Thomas Dohmke, unveiled a Git forge for hosting code, reviewing changes and running builds. Trails replaces the pull request with agentic code review.
Agent sessions sync into the repo for future use, and evals run on every push with findings in the web UI and open-source CLI. Private repos are natively hosted at launch; GitHub mirroring remains.
Source: GeekWire
Entire, founded by former GitHub CEO Thomas Dohmke, unveiled a Git forge for hosting code, reviewing changes and running builds. Trails replaces the pull request with agentic code review.
Agent sessions sync into the repo for future use, and evals run on every push with findings in the web UI and open-source CLI. Private repos are natively hosted at launch; GitHub mirroring remains.
Source: GeekWire
❤5
Junie Local swaps in Splash engine, hits ~70 tokens/s on M5 Max
Junie Local now delivers ~70 tokens/s on M5 Max for coding tasks, up from ~40 tokens/s, with bursts up to 107 tokens/s. Splash replaces MLX-VLM as its inference engine, and a DFlash2 draft model for Qwen3.8-3.6-27B-blend is included.
Faster decode and prefill change how usable a local coding agent feels on Apple silicon. Run junie update, then inside Junie run /local and press ctrl + r to update the inference engine. M4 Macs and newer with 64GB of RAM are now supported.
Source: JetBrains Junie
Junie Local now delivers ~70 tokens/s on M5 Max for coding tasks, up from ~40 tokens/s, with bursts up to 107 tokens/s. Splash replaces MLX-VLM as its inference engine, and a DFlash2 draft model for Qwen3.8-3.6-27B-blend is included.
Faster decode and prefill change how usable a local coding agent feels on Apple silicon. Run junie update, then inside Junie run /local and press ctrl + r to update the inference engine. M4 Macs and newer with 64GB of RAM are now supported.
Source: JetBrains Junie
❤1
GitHub Copilot Fusion is now generally available
Pick Copilot Fusion in the model picker instead of a single model, then choose Fusion Standard, High, or Max. It is available on all Copilot plans in VS Code, the Copilot app, Copilot CLI, and JetBrains IDEs.
Copilot routes and orchestrates models per task, so you stop switching models mid-task. Admins control which models Fusion can use and can set it as the default via Enterprise managed settings.
Source: GitHub Copilot
Pick Copilot Fusion in the model picker instead of a single model, then choose Fusion Standard, High, or Max. It is available on all Copilot plans in VS Code, the Copilot app, Copilot CLI, and JetBrains IDEs.
Copilot routes and orchestrates models per task, so you stop switching models mid-task. Admins control which models Fusion can use and can set it as the default via Enterprise managed settings.
Source: GitHub Copilot
❤1
Amp adds macOS Satellites: cloud Macs for iOS and Mac builds
Amp now has a Mac: ask it to work on your iOS or Mac app and watch it build, test, and run the app. Your thread still runs on a Linux orb; the Mac is a second machine attached to it.
Amp boots a fresh cloud Mac with Xcode and simulators in about 30 seconds, builds and tests there, and streams the screen so you can take control. macOS Satellites are in beta; join the waitlist in Settings → Experimental Features.
Source: Amp
Amp now has a Mac: ask it to work on your iOS or Mac app and watch it build, test, and run the app. Your thread still runs on a Linux orb; the Mac is a second machine attached to it.
Amp boots a fresh cloud Mac with Xcode and simulators in about 30 seconds, builds and tests there, and streams the screen so you can take control. macOS Satellites are in beta; join the waitlist in Settings → Experimental Features.
Source: Amp
❤3
SuperPlane 1.0 ships an open-source software factory for coding agents
SuperPlane 1.0 is an open source software factory that gives engineering teams a shared, repeatable workflow around their coding agents, from ticket to verified pull request, with people deciding what ships.
It is model-agnostic, so teams connect their own model accounts and API keys, and it is Apache 2.0 open source, so it can run on your own infrastructure.
Source: SD Times
SuperPlane 1.0 is an open source software factory that gives engineering teams a shared, repeatable workflow around their coding agents, from ticket to verified pull request, with people deciding what ships.
It is model-agnostic, so teams connect their own model accounts and API keys, and it is Apache 2.0 open source, so it can run on your own infrastructure.
Source: SD Times
❤3
DeepSeek prototypes Super-Agent that acts on developer chats without being prompted
DeepSeek's Harness team has unveiled an open-source "Super-Agent" prototype: it monitors developer conversations, spots coding problems and works toward fixes without waiting to be prompted.
The design splits a Router layer, which keeps broader context and handles communication, from multiple Worker sessions that run separate coding tasks in parallel. A reported bug in a group chat can be identified, reproduced and turned into a pull request without anyone tagging the agent.
Source: tradingview.com
DeepSeek's Harness team has unveiled an open-source "Super-Agent" prototype: it monitors developer conversations, spots coding problems and works toward fixes without waiting to be prompted.
The design splits a Router layer, which keeps broader context and handles communication, from multiple Worker sessions that run separate coding tasks in parallel. A reported bug in a group chat can be identified, reproduced and turned into a pull request without anyone tagging the agent.
Source: tradingview.com
❤4
LinearB Agentic Studio turns delivery data into agent context
LinearB now turns everything happening across software delivery into context that people and agents can use and act on. Agentic Studio opens on a question box, and every value comes from a live call to the LinearB public API.
Coding agents get the same history through an MCP server, plus add-ons LinearB calls agent skills. The first skill, effort, is in early access and decides how hard an agent should try per change. It works with any coding agent. Admins control access.
Source: linearb.io
LinearB now turns everything happening across software delivery into context that people and agents can use and act on. Agentic Studio opens on a question box, and every value comes from a live call to the LinearB public API.
Coding agents get the same history through an MCP server, plus add-ons LinearB calls agent skills. The first skill, effort, is in early access and decides how hard an agent should try per change. It works with any coding agent. Admins control access.
Source: linearb.io
❤1
Model mix widens: teams route across more labs and self-host more
OpenRouter data cited by Kilo says customers spending $100+ a month now use a median of 6 model families, double a year ago. Kilo's own dashboard shows the leading model changing by agent role, with no model above 30% of any role.
Developers can expect per-agent model defaults rather than one default model: cheaper open models for code and debug, frontier models for review. Kilo points to per-agent settings in Settings → Models or kilo.jsonc, local models in Kilo Desktop, and Auto Efficient routing.
Source: blog.kilo.ai
OpenRouter data cited by Kilo says customers spending $100+ a month now use a median of 6 model families, double a year ago. Kilo's own dashboard shows the leading model changing by agent role, with no model above 30% of any role.
Developers can expect per-agent model defaults rather than one default model: cheaper open models for code and debug, frontier models for review. Kilo points to per-agent settings in Settings → Models or kilo.jsonc, local models in Kilo Desktop, and Auto Efficient routing.
Source: blog.kilo.ai
AxiomStudio.ai launches Vera, an AI code review agent for every pull request
Vera reviews pull requests on GitHub, GitLab, Azure DevOps, and Bitbucket, flagging bugs, logic errors, security risks, and maintainability issues before merge. It is included on all AxiomStudio tiers with unlimited reviews and users.
Findings can become tracked work in Jira, Linear, ClickUp, Inflectra, Azure DevOps, or GitHub Issues in one click. Vera works with Codex, Claude, Gemini, Qwen, and Cursor on existing plans, with no extra LLM subscription.
Source: businessinsider.com
Vera reviews pull requests on GitHub, GitLab, Azure DevOps, and Bitbucket, flagging bugs, logic errors, security risks, and maintainability issues before merge. It is included on all AxiomStudio tiers with unlimited reviews and users.
Findings can become tracked work in Jira, Linear, ClickUp, Inflectra, Azure DevOps, or GitHub Issues in one click. Vera works with Codex, Claude, Gemini, Qwen, and Cursor on existing plans, with no extra LLM subscription.
Source: businessinsider.com
❤1
AWS launches Kiro bots in Kiro Crew for parallel coding tasks
AWS launched Kiro bots, generally available within Kiro Crew, its open-source multi-agent development environment, to help developers coordinate parallel coding tasks, maintain context across sessions, and automate recurring workflows.
Create a bot via Customize and "Add a bot," based on a default or existing custom agent. Each bot runs multiple worker-agent sessions in parallel, monitors and redirects them, consolidates results, and keeps persistent memory via Notes.
Source: InfoWorld
AWS launched Kiro bots, generally available within Kiro Crew, its open-source multi-agent development environment, to help developers coordinate parallel coding tasks, maintain context across sessions, and automate recurring workflows.
Create a bot via Customize and "Add a bot," based on a default or existing custom agent. Each bot runs multiple worker-agent sessions in parallel, monitors and redirects them, consolidates results, and keeps persistent memory via Notes.
Source: InfoWorld
❤1