Gemini's New Version Tops Arena Leaderboard, But Is Jailbroken Shortly After Release
Multimodal failures and model personality issues: Interaction styles during code debugging and security jailbreak cases spark reflections on product reliability.
Frontier Models
JustGhostIt — Frontier Models archive, page 3.
Filter by category
Google's Annual Showcase: All AI Models Upgraded; Gemini 2.5 Dominates Top Spots, New Video/Image Models Debut
At Google I/O, the entire Gemini suite was upgraded. Gemini 2.5 lines, video (Veo), image models, and developer APIs received high-density updates on the same day.
Google's Annual Power Move: All AI Models Upgraded; Gemini 2.5 Dominates Rankings, New Video/Image Models Debut
Gemini 2.5 family GA and Flash-Lite introduce segmented SKUs along the latency-cost curve to target enterprise and edge scenarios.
GPT-4V Only Reaches Level-2? Global First Multimodal Generalist Ranking Released, General-Level Sets New Paradigm for Evaluating Multimodal AI
Anthropic releases Claude 4 (Opus/Sonnet), focusing on long-horizon autonomous tasks and coding scenarios, while simultaneously raising API pricing tiers.
Large Models Fail en masse! New Chinese Web Search Test Shows GPT-4o Accuracy at Just 6.2%
Search experience revolution: 'AI Mode' expands in the US, coexisting with traditional link lists, sparking debate over traffic distribution and ad landscapes.
Testing o3/o4-mini: Solving Euler's Problem in 3 Minutes Proves OpenAI's Top Model Lives Up to Its Reputation
OpenAI releases reasoning models like o3 and o4-mini, emphasizing tool use and scientific/programming logic. Third-party reproductions and hallucination rates spark debate.
Toward Swarm Intelligence: BAAI Unveils First Cross-Embodiment Brain-Cerebellum Collaboration Framework and Open-Source Embodied AI Brain
BAAI releases a robot brain-cerebellum collaboration framework and open-source ecosystem, advancing real-world data integration and swarm intelligence narratives.
Google Unveils Gemini 2.5 AI Model
Google enhances Gemini 2.x with advanced reasoning and search integration, iterating on its 'thinking' multimodal model while expanding generative capabilities across search, developer tools, and cloud services.
Lightelligence Unveils World's First Next-Gen Photonic-Electronic Hybrid Computing Card
Lightelligence reveals new photonic-electronic hybrid computing hardware, aiming to overcome energy efficiency bottlenecks in matrix operations for AI.
11B Model Achieves SOTA in Open-Source Video Generation with Just 224 GPUs, Cutting Training Costs by 10x
Open-source video generation breakthrough: Domestic teams and the global open-source community advance billion-parameter models and low-cost training, narrowing the gap with closed commercial solutions.