Harbor celebrates its first birthday and reports processing 200T tokens per week, over 1B rollouts, and 7.3M PyPI downloads. It also notes 10M+ Terminal-Bench runs and 100T+ tokens/week through Harbor.
Key points
Tokens processed: 200T tokens per week now
Rollouts: 1B+ Harbor rollouts
Sources
Moneta — Your work is the content launches for $20 one-time
Moneta — Your work is the content launches a Mac app that captures a screenshot each minute and lets a local agent draft public posts, priced at $20 one-time. It runs locally on macOS 14 or later and stores screenshots and logs on the Mac.
Key points
$20 one-time purchase, tax included, licence key works on up to 3 Macs.
Captures a screenshot every minute, keeps files locally, only contacts server to verify licence weekly and check updates daily.
Sources
- Original post
- Moneta — Your work is the content - Linked page
dreamDEXSomnia on Somnia Network 3ms onchain execution
dreamDEXSomnia on Somnia Network shows 3ms fully onchain execution in Tokyo with zero trading fees. The post highlights rising competition in onchain perpetuals.
Key points
Execution latency: 3ms fully onchain execution in Tokyo
Trading fees: zero trading fees
Sources
- Original post
- help.x.com - Linked in the post
OpenAI models turn-based vs continuous listening claim, 20 ms response chunks
OpenAI former CTO Mira Murati claims current models are turn-based, waiting for input then thinking while deaf and blind. She says next models will continuously ingest audio, text, video and answer in 20-millisecond chunks.
Key points
Current models are turn-based, pausing to think after user input.
Future models will listen continuously and respond in 20-millisecond chunks.
