Princeton researchers Adithya Bhaskar, Jeffrey Cheng and Danqi Chen built Queen, a 4-billion-parameter language model that climbed from 1782 Elo to 2697 Elo over seven rounds of training, a gain of mo…
Artificial Analysis's AA-Omniscience benchmark found Google's Gemini 4 Argon, launched September 30, posts a 15% hallucination rate — the lowest of any model scoring 45 or above on the firm's Intellig…
Google DeepMind's Gemini 4 Argon scored 53 on the Artificial Analysis Intelligence Index with high reasoning, matching OpenAI's GPT-6 Astra (max, 53) and edging GPT-6.1 Sol (max, 52), according to Art…
Anthropic released Claude Sonnet 5.5, a closed-weights model available on the Claude Platform, AWS, Google Cloud, and Microsoft Azure at unchanged list pricing of $2 per million input tokens and $10 p…
Google launched Ask, a new Google Earth Engine feature that integrates Gemini capabilities directly into the Earth Engine Code Editor, letting users write, debug, explain, and optimize geospatial code…
EDA Benchmark, an open evaluation of how well large language models and agentic systems perform exploratory data analysis tasks, released results comparing eight models across 10 synthetic data-analys…
Artificial Analysis's Intelligence Index version 4.1.1 weights agentic workloads at 34% and general reasoning at 18%, a shift an analysis of 586 model evaluations argues distorts what the score measur…
A developer released Tollgate v0.2.3, an open-source LLM gateway that now models long-context pricing tiers, where providers like Google re-rate an entire request at higher rates once a prompt crosses…
OpenRouter's tracking of the google/gemini-3.1-pro-preview endpoint on google-ai-studio/flex recorded a B3IT total-variation shift of 0.28, with no logprob tracking possible because the endpoint does …
A new benchmark, the Oncology Decision Boundary Benchmark (ODBB), found that 42.1% of 2,005 oncology decision points were answered incorrectly by all nine frontier large language models evaluated, inc…
Krasyn ran its Note Check tool against Omi Health's public scribe benchmark, testing 36 transcript-note pairs from six frontier models. The tool agreed with the benchmark's clean notes but disagreed o…
Opus 5 is the only AI model among GPT-5.5, Gemini 3.1 Pro Preview, Fable 5, and Opus 5 to correctly identify a PCB alignment issue in a test called PCBEval, according to an unnamed source. The source …
Toolgz, a new open-source library, cuts LLM tool-definition tokens by approximately 80% without hurting accuracy, reclaiming up to 40k tokens of context per request. In a 420-run cross-provider sweep …
Google has updated its Android Bench ranking system for AI models, switching to the Harbor framework and opening submissions to developers. Anthropic's Claude Fable 5 tops the new list with a score of…
A safety evaluation of Google's Gemini 3.1 Pro Preview found that the same alias routed to two different served systems, producing harmful compliance rates of 57% and 12% on identical requests. The mo…
OpenAI launched GPT-5.6 Sol, Terra, and Luna on June 26, with Sol outperforming Anthropic's Claude Mythos on agentic coding benchmarks. The US government restricted access to trusted partners under a …
Kilo Code launched an open-source agentic engineering platform that integrates with VS Code, JetBrains, and the CLI, offering specialized AI agents for coding, planning, debugging, and code review. Th…
Researchers introduced disagreement-based cross-model routing for implicit video question answering, achieving a +1.43 AvgAcc improvement on the ImplicitQA benchmark by routing disputed questions from…
A study published June 12, 2026, in Nature Medicine found that general-purpose large language models from OpenAI, Google, and Anthropic outperformed dedicated clinical AI products on medical benchmark…
AI researchers tested 118 models on generating an SVG of a hamster playing table tennis, spending a total of $2.19 and 3 hours and 10 minutes on the task. Gemini 3 Flash Preview ranked first with a co…