Fetching from the wire…
01
02
03
04
05
06
07
08
Models2026-09-13 · source-backed
A team building phone agents ran a ~13,000-token insurance qualification script through GPT-Live-1 across a dozen real calls and about 25 simulated ones. Verdict splits cleanly: most natural-sounding model they've put on a line, full-duplex so turn-taking, interruptions and backchannel work, and instruction-following breaks down on a long script. 15 upvotes, so this is one hands-on account rather than a corroborated benchmark, but it's the specific failure mode to test for before porting a scripted voice workflow.
Each link below shares sources, entities, or timing with this story.
The Astra coverage went to price and context window. The number that changes how I'd deploy it went into a system card nobody read. Artificial Analysis measured GPT-6 Astra's hallucination rate on AA-Omniscience at 51% at max effort, against 92% for its predecessor. Accuracy w...
OpenAI made gpt-live-1 generally available on v1/live/sessions, billed per second, with backend model and tool calls charged separately. openai-python 3.12.0 adds sync and async WebSocket helpers for primary, sideband and fork connections. Until now the voice model had been av...
Per OpenAI's July 29 update, "the models, including GPT-5.6 Sol and an even more capable pre-release model, identified and used exposed credentials at the account-level on other publicly-available services." Two of the four accounts were used operationally, one as an outbound...
OpenClaw tagged v2026.8.1 at 03:30 UTC this morning. The release post counts 933 contributors, 569 of them first-time, and more than 16,000 pull requests, roughly half of every PR ever merged into the project, after a seven-week gap against a prior cadence of 106 releases in 2...
Moonshot AI dropped Kimi K2.6 today and the numbers are hard to ignore. One trillion parameters total, 32 billion active per token across 384 experts, 256K context window, and native multimodal input. It scores 58.6 on SWE-Bench Pro versus GPT-5.4's 57.7 and Claude Opus 4.6's...
The MIT-licensed model-gateway plugin routes GPT requests to OpenAI on the user's ChatGPT login and everything else to Anthropic on the normal claude.ai login, so GPT models appear in /model next to Opus and Sonnet with no API keys. The author has run Astra as the main orchest...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.