Meta researchers released SWE-Together, a benchmark built from 11,260 real developer-agent sessions that tests AI coding agents on collaborative performance under user corrections, not just one-shot t…
GitHub will deprecate Gemini 2.5 Pro and Gemini 3 Flash across all Copilot experiences on July 31, 2026, recommending Gemini 3.1 Pro and Gemini 3.5 Flash as alternatives. Users must update workflows a…
Goldman Sachs analyst Rich Privorotsky projects hyperscaler capital expenditures could reach $765 billion by 2026 and cumulative AI infrastructure spending $7.6 trillion by 2031, driven by strong dema…
OpenRouter launched Fusion API on June 13, an API that routes prompts to multiple AI models in parallel and synthesizes outputs, days after the U.S. government forced Anthropic to pull its Fable 5 and…
Goldman Sachs' 1-Delta Desk, led by Rich Privorotsky, flagged a chart tracking open-source vs. closed-source AI models as a leading indicator for the AI infrastructure trade, arguing that AI is shifti…
GLM 5.2, a new open-weights model, achieved 15% fewer achievements than Gemini 3 Flash in text adventure games, a statistically significant difference. The benchmark, costing $5.1, controlled for game…
TrustedRouter achieved a state-of-the-art score of 70.6 on the DRACO benchmark by fusing five models including open-weights models DeepSeek V4 Pro and Kimi K2.6, surpassing OpenRouter's best fusion of…
A new study reveals that large language models exhibit a strong brand bias, recommending well-known brands 100% of the time when product specifications are identical, but this dominance can be broken …
A new study finds that online web agents augmented with memory, workflow, or skill modules often fail to outperform a token-matched vanilla baseline under a fixed inference budget. Testing across thre…
Google DeepMind researchers trained Gemini 3 Flash to exhibit positive traits by midtraining on synthetic documents describing the model's traits, then finetuning on synthetic chat data where it demon…
A test of Google's Gemini 3 Flash Preview found that 67% of AI-generated curl commands were unsafe, targeting internal networks, cloud metadata endpoints, or localhost. The test, conducted by Check, g…
Google DeepMind researchers investigate why filtering supervised fine-tuning (SFT) data fails to remove safety-relevant properties from language models, proposing a method to identify the source of th…
OpenRouter launched Fusion, a tool that synthesizes outputs from multiple AI models into a single response, achieving scores of 69.0% on the DRACO deep research benchmark—surpassing individual frontie…
Google DeepMind researchers found that supervised fine-tuning (SFT), not reinforcement learning, drives most safety properties in Gemini models. Comparing SFT-only versions of Gemini 3.1 Pro and Gemin…
A team at Redwood Research developed a proof-of-concept pipeline that transforms benign Claude Code transcripts into synthetic sabotage trajectories for automated red-teaming of AI monitors. The appro…
Google has released Gemini 3.5 Flash, an AI model designed to close the gap between speed and intelligence, particularly for complex, multi-step agentic workflows. The model features a 1 million token…