Fresh stories
Nous releases Arete-27B with an 803-prompt alignment evaluation
Nous released Qwen-based Arete-27B, training data and an 803-prompt evaluation. It reports cutting its measured propaganda rate from 22% to 1% without a significant Hermes Bench drop.
MaxxBench tests coding models on altered DeepSWE requirements
MaxxBench changes details such as filename casing and timestamp formats in DeepSWE tasks. Critics dispute whether failures establish memorization, citing task selection and unclear failure labels.
OpenAI says Codex Pro plans remove five-hour usage limits
OpenAI says Pro 100, 200 and 500 offer 10×, 20× and 50× Plus usage without five-hour limits. It also says allowances doubled for Pro plans and Business 100.
Nous releases Arete-27B with an 803-prompt alignment evaluation
Nous released Qwen-based Arete-27B, training data and an 803-prompt evaluation. It reports cutting its measured propaganda rate from 22% to 1% without a significant Hermes Bench drop.
ReproBridge paper reports experiment reproduction rising to 92.6%
A summary of Google’s ReproBridge paper reports reproduction success rising from 76.4% to 92.6%, with API spending down 19.7%. It packages code, environments and launch scripts for coding agents.
Top storiesthis week
Inkling 2 Small scores 81.7% on ARC-AGI-2 at $0.27 per task
ARC Prize verified Inkling 2 Small at 81.7% on ARC-AGI-2 for $0.27 per task and 96.5% on ARC-AGI-1. Letta separately reports 60% on FrontierMemory; full weights are planned for next week.
OpenRouter Auto Code claims Opus-level coding at 74% lower cost
OpenRouter launched Auto Code, reporting Opus-level coding performance at 74% lower cost. Its routers explain decisions and adapt to model restrictions; beta reportedly cut invalid tool calls fourfold.
Meta adds Active Memory models to Muse and Meta AI
Meta is deploying specialized models with Active Memory and Active Focus in Muse and Meta AI. It reports 2.4× less forgetting of critical information and a 67% improvement in recovering from unexpected obstacles.
Nace AI releases Drex models that output calibrated probabilities
Nace AI released Drex, an open-weight multimodal decision-model family that outputs calibrated probabilities. Its training report describes verified decision training, while reports put the 27B Max response time at 183 ms.
Vals reports DeepSeek completed 72.8% of simulated cyber objectives
Vals reports DeepSeek completed 72.8% of objectives in isolated enterprise cyber simulations, versus 80.8% for GPT-6 Sol through Daybreak. On Lighthouse, DeepSeek completed 21 objectives at about a tenth of the cost.