AI Agents
AI Agent technology and applications
Articles
Perplexity Trains Its Computer Agent on Real Mistakes With Hint⭐9
Perplexity trains its computer agent on real mistakes using hint-guided self-distillation, cutting tool-call failures 21.2% in live testing.
BottleCap AI Releases ThinkingCap-Qwen3.8-27B: 37.2% Fewer⭐8
ThinkingCap-Qwen3.8-27B by BottleCap AI cuts thinking tokens by 37.2% on average across 12 benchmarks, with only a 0.86pp accuracy drop.
A Coding Guide to TypeSafe AI Jev: Typed Decisions, Calibrated⭐9
Learn to build TypeSafe AI Jev with typed decisions, calibrated confidence, and speculative fan-out using a System One model in this hands-on coding guide.
Google Releases Gemini 3.8 Flash TTS and Flash-Lite TTS With⭐8
Google releases Gemini 3.8 Flash TTS and Flash-Lite TTS, bringing prompt-based voice design and expressive audio generation to the Gemini API.
Kyutai Releases Voice of Reason: A Speech-Native Model that⭐9
Kyutai's Voice of Reason uses reinforcement learning to lift GLM-4-Voice from 27.3% to 77.1% on spoken GSM8K math.
Nokia Open-Sources AnyJev: A Training-Free Layer That Turns Any⭐9
Nokia's open-source AnyJev turns open LLMs into calibrated decision models with no training, lifting automatable traffic 6.8x
SpeakON Ships a MagSafe AI Voice Button With Its Own Microphone⭐8
SpeakON launches a MagSafe AI voice button with its own microphone that turns speech into finished text in any open app, keeping your iPhone mic free.
OpenAI Releases GPT-6 Sol and Luna: 50% Cheaper API Pricing and⭐8
OpenAI launches GPT-6 Sol and Luna with 50% cheaper API pricing. See benchmarks, tier comparisons, and how they stack up against Astra and Claude Opus 5.
SpeakON Launches MagSafe AI Voice Button with Dedicated⭐9
SpeakON launches a MagSafe AI voice button with its own mic that turns speech into polished text and cross-app actions, not raw dictation.
Anthropic Releases Claude Opus 5.5: Fable 5.1-Level Performance⭐8
Anthropic's Claude Opus 5.5 delivers Fable 5.1-level performance at 40% lower running cost than Opus 5, with benchmark leads in coding, computer use, and knowle...