AREX-2, an LLM agent built on Qwen3.8-27B and trained on synthesized long-horizon improvement trajectories from machine learning and algorithmic programming tasks, scored 81.8 on MLE-bench Lite and 70…
A 35B-parameter model trained with a three-state autonomous search harness (Rubric, Answer, Verify) and a Seal Memory tool scored 72.83 on BrowseComp, outperforming comparable open-source systems, acc…
Exa released Agent Ultra, the highest effort level of its Exa Agent API, which the company says beats Opus 5.5, GPT-6 Astra, and Perplexity Agent at maximum effort on 4 research benchmarks, including …
Iris-pro, a 397B parameter search agent developed by an open-source research team, achieved a 92.9 accuracy score on the DeepSearchQA benchmark with context management enabled, setting a new state-of-…
MiniMax M3 Medium, developed by MiniMax and accessed via OpenRouter, achieved a 73.17% F1 score on the DeepSearchQA benchmark, approaching the performance of GPT-5 High. The model completed 8 of 9 tri…
Artificial Analysis released DeepSearchQA, a benchmark of 900 tasks for evaluating LLM search API providers, where an LLM grader scores answers using F1 over answer items for broad research questions …
A developer built Khidi, an API gateway that makes cheap open-weight AI models as easy to use as expensive proprietary ones by handling the plumbing of inference sourcing, key management, and billing.…