Command Code Adds Muse Spark 1.3, Its Cheapest Model
Command Code has added Muse Spark 1.3, its cheapest model option, with a tweet referencing "Flappy bench" as a performance indicator.
Benchmarks Are Telling You: What You Think Matters Doesn't Matter
A developer building AI coding tools argues that benchmark rankings are misleading—the metrics people obsess over don't reflect what actually drives adoption an…
Claude Code Keeps Topping the Charts Despite Losing Every Benchmark
A developer notes that Claude Code keeps ranking last in new benchmarks, yet it remains the #1 AI coding tool and is still growing faster than most rivals — a p…
OpenCode Says It Won't Train Its Own Model — For Now
OpenCode's builder explains why the platform is staying out of the model-training race: capability gaps, crowded competition, and a commitment to being neutral…
Microduck Meets Reachy Mini: Hugging Face Asks the Community for Robot Scenarios
Hugging Face highlighted a playful clip of Microduck discovering Reachy Mini and invited the community to suggest fun scenarios for the two robots.
Meta Muse Spark 1.3 Now Free on OpenCode
Meta's Muse Spark 1.3 model is now freely available on OpenCode, bringing frontier-level capability to the open-source coding agent.
Muse Spark 1.3 Reminds Developer of the Value of Fast AI Models
A developer using Muse Spark 1.3 notes how much they missed fast models, highlighting the practical value of speed in AI-assisted coding workflows.
Gemini 3.8 Flash Now Available on GMI Cloud for Agentic Workloads
GMI Cloud announces Gemini 3.8 Flash availability, touted as a well-rounded agentic model, accessible via GMI API key for agent harnesses.
GMI Cloud and MiniMax Host Live Session on M3 for Agents, Coding, and Reasoning
GMI Cloud and MiniMax are hosting a live session on the M3 model for agents, coding, and reasoning, including a live build with a GMI API key.
LongCat-2.0 Now Free in Command Code: 1.6T Parameters, 1M Context
Command Code has made LongCat-2.0 free for all subscribers on every plan, featuring 1.6T total parameters, 48B active parameters, and a 1M context window.