#LLM
Curated coverage on «LLM»: releases, research, benchmarks, open-source.
OpenAI Releases Decisions API: Agent Decisions Up to 10x Faster
Decisions API on GPT-6 Luna exits beta: up to 10x faster than Responses API, $0.10 per million input tokens, zero charge for output. Sierra hits 280 ms p95.
Meta releases Muse Spark Open: 816B parameters and the lead among US open-weight models
Muse Spark Open, with 816B parameters and 42B active, is the strongest US open-weight model and has halved the gap between US and Chinese open-weight models.
SLICE-Mem Cuts AI Agent Memory Token Costs by 92% Without Sacrificing Quality
Researchers at Hong Kong Polytechnic University have found a way to store AI agent memories without constant calls to a large language model — accuracy barely…
aiedit hides individual lines of code from an AI agent: 5 levels and 857 tests
vps-forge: Agents With Root on a VPS Get Auto-Rollback, Self-Bans Drop From 7 in 60 Runs to 0
BigSleep LLM Agent Finds 45 V8 Vulnerabilities, Fuzzilli 37, With 25% Overlap
rwkv-quant: keeping one matrix in bf16 removes 39% of RWKV-7 2.9B's quantization loss
OpenAI Releases GPT-6.1 Terra: Half the Price, Twice the Accuracy in Code
GPT-6.1 Terra solves 50.3% of Terminal-Bench 4.0 tasks versus 21.5% for GPT-5.6 Terra. The API costs $1 per million input tokens, 50% less than Sol and the previous…
Qwen3-8B on the Intel Core Ultra 5 226V NPU: 10.7 tok/s at 15 W and 24 dB of Noise
Qwen3-8B on all three Lunar Lake blocks in a Ninkear L20 mini PC: the NPU hits 10.7 tok/s at 1.52 J/token, the GPU is twice as fast, and the machine stays quiet…
Microsoft and Hugging Face trained Qwen3.8-27B on database-state rewards: from 38 to 107 tasks
Gemma 26B fine-tuned on a 24 GB MacBook: offline assistant gets 703 of 766 right
Cognition releases FrontierCode 2.0: the best score is just 45.5%
Dharma-AI trimmed Qwen3-4B's vocabulary and cut inference cost by 77%
Meta open-sources Muse Spark Open: an 816B-parameter MoE with a 1M-token context
Meta has released Muse Spark Open, an open-weight model with 816B parameters, 42B active per token, text, image, audio and video input, and tuning for agentic…
Six AI Agents Forecast Earnings Per Share: Three Matched the Analysts
Georgia Tech researchers tested six LLM agents on 149 quarterly reports from 86 US companies. Three matched the analyst consensus, but all underestimate earnings…