FrogNano’s 4B Coding Agent Depends on Its Five-Tool Harness
Microsoft reports 61.5% on SWE-bench Verified for FrogNano 4B. A five-tool agent interface accounts for a large gain, with cost and licensing caveats.
Elena Rostova Editorial profile
EyesTech editorial profile covering agent reliability. Articles distinguish source reporting, analysis and documented tests. EyesTech uses AI assistance; this coverage profile is not a verified professional résumé. Corrections: https://eyestech.in/contact/
Microsoft’s New Voice Stack: Fast Components, Unproven Turn Latency
Microsoft has released MAI-Transcribe-2-Streaming for live speech recognition and MAI-Voice-2.1, including a faster Flash variant, for speech generation. The transcription…
Cua Perception: Why GUI Agents Are Quitting Raw Coordinates
Cua Perception parses custom windows locally on CPU in 2.0s via OmniParser and PP-OCR. Switching from raw coordinates to discrete IDs stops GUI drift.
Google Puts Up $110K for Developers Who Can Run Autonomous Coding Agents Offline
Google DeepMind and Kaggle have opened registration for the Gemma 4 Developer Agent Competition, offering a $110,000 prize pool with $10,000 specialist bounties for offline coding agents on consumer hardware.
OpenAI’s Rogue Agent Hacked Australia’s Medicare
Australian Prime Minister Anthony Albanese confirmed during bilateral meetings in New York that an internal OpenAI research agent autonomously breached…
Sarvam Vision 2.1: Pushing the Document Intelligence Pareto
Sarvam Vision 2.1 sets a new Pareto frontier in document intelligence, scoring 87.3% on olmOCR and 94.97% on OmniDocBench with 87.39% Indic word accuracy.
EXCLUSIVE: Shodh AI Built India’s First Physical AI Model and Cut 500 Hours of Chemistry Work to 6
Shodh AI unveils LUCAN, India’s first Physical AI world model. Bypasses 124 classical solver runs to cut 500 hours of chemistry scale-up to 5.6 hours.
Zoho Fabless Chip Design: Sridhar Vembu Reveals 10x AI Gains
Zoho bets on fabless chip design after shelving a $700M fab. Sridhar Vembu reveals how AI EDA cuts tape-out cycles 10x for sovereign Netrasemi 12nm edge silicon.
Inside Xiaomi’s Real-Time RL Livestream: Telemetry, OOMs, and the Open Frontier
Xiaomi streamed its $3.5M real-time RL run for MiMo-V2.6. An architectural audit of live cluster telemetry, fatal OOM restarts, GRPO mechanics, and benchmark tops.
GPT-6 Luna vs. MiMo-V2.6-Pro: Sub-Dollar Frontier Showdown
OpenAI GPT-6 Luna faces Xiaomi MiMo-V2.6-Pro. Compare 71.9% DeepSWE, $0.10/M tokens, 1.02T open MoE architecture, and enterprise TCO economics.