#DeepSearchQA
LightNews LightNews
gigazine.net
Googleが「Gemini Deep Research エージェント」をリリース＆ベンチマーク「DeepSearchQA」オープンソース化
https://gigazine.net/news/20251212-gemini-deep-research/
Googleが「Gemini Deep Research エージェント」をリリース＆ベンチマーク「DeepSearchQA」オープンソース化
Googleが、「Gemini Deep Research」のウェブ検索能力を大幅に強化した「Gemini Deep Research エージェント」を、Interaction API経由で利用可能にすることを発表しました。また、ディープサーチエージェント評価用のベンチマーク「DeepSearchQA」のオープンソース化も行われています。
gigazine.net
December 12, 2025 at 3:07 AM
technews4869.bsky.social
<a href="https://gigazine.net/news/20251212-gemini-deep-research/" class="hover:underline text-blue-600 dark:text-sky-400 no-card-link" target="_blank" rel="noopener" data-link="bsky">gigazine.net/news/20251212...
Googleが「Gemini Deep Research エージェント」をリリース＆ベンチマーク「DeepSearchQA」オープンソース化
Googleが「Gemini Deep Research エージェント」をリリース＆ベンチマーク「DeepSearchQA」オープンソース化
gigazine.net
December 14, 2025 at 8:50 AM
timkellogg.me
comparing to gpt-oss:120b, looks like shimmer is the better agentic model, but gpt-oss is better at traditional LLM tasks
Glimmer's published numbers include: • SWE-Bench Verified: 76.0 • SWE-Bench Pro: 51.2 • TerminalBench 2.1: 51.7 • MCP-Atlas: 75.5 • DeepSearchQA: 74.6 • GPQA Diamond: 83.5 • HLE Text: 22.0 • AIME 2026: 94.7 OpenAl reports gpt-oss-120b at: • SWE-Bench Verified: 62.4 • GPQA Diamond: 80.1, or 80.9 with tools • HLE: 14.9, or 19.0 with tools • AIME 2025: 92.5, or 97.9 with tools • Codeforces: 2463, or 2622 with terminal access
August 10, 2026 at 2:15 PM
kaggle.com
🚀 New on Kaggle Benchmarks: DeepSearchQA developed by Google DeepMind!

This benchmark focuses on complex web research tasks and tests agent comprehensiveness.

Check the leaderboard: www.kaggle.com/benchmarks/g...
A screenshot of the Kaggle DeepSearchQA leaderboard, showing the top five ranked models.
December 11, 2025 at 6:30 PM
bild.bsky.social
Google DeepMind, geliştiricilerin yeni Interactions API üzerinden erişebileceği geliştirilmiş Cem Ceminay (Gemini) Deep Research ajanı ve yeni DeepSearchQA benchmarkını yayınladı.
Build with Gemini Deep Research
We have reimagined Gemini Deep Research to be more powerful than ever, now accessible to developers via the new Interactions API.
blog.google
December 12, 2025 at 8:56 AM
winbuzzer.com
winbuzzer.com/2025/12/11/g...

Google Releases Gemini 3 Powered Deep Research Agent to Developers via API

#ArtificialIntelligence #GenerativeAI #Google #GoogleAI #AgenticAI #DeepResearch #Gemini3 #DevTools #API #TechNews #Winbuzzer #LLM #MachineLearning #DeepSearchQA #OpenSource
Google Releases Gemini 3 Powered Deep Research Agent to Developers via API - WinBuzzer
Google has released its Gemini Deep Research agent to developers via the Interactions API, offering a low-cost, stateful alternative to OpenAI's tools.
winbuzzer.com
December 11, 2025 at 6:46 PM
tech-trending.bsky.social
Googleが「Gemini Deep Research エージェント」をリリース＆ベンチマーク「DeepSearchQA」オープンソース化
https://gigazine.net/news/20251212-gemini-deep-research/
Googleが「Gemini Deep Research エージェント」をリリース＆ベンチマーク「DeepSearchQA」オープンソース化
Googleが、「Gemini Deep Research」のウェブ検索能力を大幅に強化した「Gemini Deep Research エージェント」を、Interaction API経由で利用可能にすることを発表しました。また、ディープサーチエージェント評価用のベンチマーク「DeepSearchQA」のオープンソース化も行われています。
gigazine.net
December 12, 2025 at 4:40 PM
youshenlim.bsky.social
#Google launched Deep Research and Deep Research Max: autonomous agents that blend open web and private data via MCP in one API call.

Built on Gemini 3.1 Pro. Max hits 93.3% on DeepSearchQA.

Caveats: API only, 60-min limit, MCP quality depends on the server, benchmarks ≠ real research.
Deep Research Max: a step change for autonomous research agents
Introducing Deep Research and Deep Research Max, the next generation of Google’s autonomous research agents.
blog.google
April 22, 2026 at 9:19 PM
kimi-moonshot.xmirror.bot
🥝 Meet Kimi K2.5, Open-Source Visual Agentic Intelligence. (1/4)
The image is a bar chart comparing the performance of Kimi K2.5, GPT-5.2 (xhigh), Claude Opus 4.5, and Gemini 3 Pro across various benchmarks related to agents, coding, image, and video tasks. Each benchmark is represented with a percentage score, with the tallest bars indicating the best performance. The comparative metrics include Humanity's Last Exam (HLE), BrowseComp, DeepSearchQA, SWE-bench Verified, SWE-bench Multilingual, MMMU Pro, MathVision, OmniDocBench 1.5*, VideoMMMU, and LongVideoBench. Kimi K2.5 shows strong performance in most categories, particularly in image and video processing tasks. The percentages indicate the percentile scores achieved by each model, with a higher score denoting superior accuracy.
January 27, 2026 at 6:48 PM
druce.ai
Google launched Deep Research and Deep Research Max, Gemini 3.1 Pro-powered AI agents that combine web search and proprietary enterprise data via the Model Context Protocol, generate native charts, and achieved 93.3% on DeepSearchQA while targeting finance and life sciences.
Google’s new Deep Research and Deep Research Max agents can search the web and your private data
Google unveiled Deep Research and Deep Research Max, new Gemini 3.1 Pro-powered AI agents that combine web search, proprietary enterprise data, MCP integrations, and native charts to automate high-stakes research workflows in finance, life sciences, and market intelligence.
venturebeat.com
April 21, 2026 at 11:22 PM
riccinov.bsky.social
The harness, not the model, may be the lever. JIT-Agent synthesizes task-specific agent scaffolds on the fly: DeepSeek-V4-Flash passes GPT-5.6 by 9.1 points on DeepSearchQA and 4.3 on OdysseyBench, while GLM-5.2 gains up to 20.2. Scaffolding is becoming a trainable layer.
August 28, 2026 at 2:29 AM
k-tai.watch.impress.co.jp.web.brid.gy
Google、「Gemini Deep Research」をアプリに組み込めるAPI公開
米Google（グーグル）は、「Gemini Deep Research エージェント」と、これを開発者が利用するためのAPI「Interactions API」を公開した。あわせて、複雑な検索タスクを評価するベンチマーク「DeepSearchQA」をオープンソース化して提供する。
k-tai.watch.impress.co.jp
December 12, 2025 at 10:11 PM
techdefused.bsky.social
Google launches autonomous research agents for finance, life sciences and market intelligence
Google launches autonomous research agents for finance, life sciences and market intelligence
Deep Research Max scores 93.3% on DeepSearchQA, up from 66.1% in December, as the company targets enterprise workflows
dlvr.it
April 22, 2026 at 4:37 PM
dausnart.com
Google has also open-sourced DeepSearchQA, a new benchmark for evaluating multi-step research agents.
December 15, 2025 at 8:26 PM
infodocket.bsky.social
Google Releases a "More Powerful" Deep Research Agent & Open Sources a New Web Research Agent Benchmark, #DeepSearchQA blog.google/innovation-a... & Gemini 3 Flash Comes to Gemini App blog.google/products-and... #AI #LLMs #GenAI #deepresearch
January 9, 2026 at 2:03 PM
lunamoth.bsky.social
- 구글이 Interactions API를 통해 제미나이 딥 리서치를 출시하여 개발자가 고급 자율 연구 기능을 애플리케이션에 임베드할 수 있게 함

- 제미나이 3 프로 기반으로 반복적으로 조사를 계획하고 쿼리를 생성하며 지식 격차를 식별함

- HLE 에서 46.4%, DeepSearchQA 벤치마크에서 66.1% 달성

오늘 보니 제미나이 공웹 빠른 모드로 하니 딥리서치 생기더군요. (원래 있었던 것 같긴합니다만)

Build with Gemini Deep Research

blog.google/technology/d...
Build with Gemini Deep Research
We have reimagined Gemini Deep Research to be more powerful than ever, now accessible to developers via the new Interactions API.
blog.google
December 12, 2025 at 12:38 PM
vanuatutech.bsky.social
Google DeepMind launches an enhanced Gemini Deep Research agent accessible to developers via its new Interactions API, along with a new DeepSearchQA benchmark (The Keyword)
The Keyword: Google DeepMind launches an enhanced Gemini Deep Research agent accessible to developers via its new Interactions API, along with a new DeepSearchQA benchmark  —  We have reimagined Gemini Deep Research to be more powerful than ever.  It is now accessible to developers via the new Interactions API …
www.techmeme.com
December 11, 2025 at 5:45 PM
cryptoknightai.bsky.social
Kimi正式开源旗舰模型K2.6，在代码能力、长程任务执行和Agent集群协作上实现重大突破，核心亮点如下：

1. 性能全面领先
•基准测试：在博士级考试Humanity’s Last Exam（54.0%）、Agent检索DeepSearchQA（92.5%）、软件工程SWE-Bench Pro（58.6%）等评测中超越GPT-5.4、Gemini 3.1 Pro等主流模型。
•短板：多语言编程（SWE-bench）、复杂工具调度（Toolathlon）及视觉任务（MathVision）仍小幅落后顶尖模型
April 21, 2026 at 8:57 AM
techmeme.com
Google DeepMind launches an enhanced Gemini Deep Research agent accessible to developers via its new Interactions API, along with a new DeepSearchQA benchmark (The Keyword)

Main Link | Techmeme Permalink
December 11, 2025 at 5:40 PM
longtail-news.bsky.social
Feed: "GadgetBond"
By: Shubham Sawarkar on Thursday, April 23, 2026
Kimi K2.6 is Moonshot’s new engine for autonomous coding and research
Moonshot AI’s Kimi K2.6 open-source release focuses on long-horizon coding, 300-agent swarms, and real benchmark gains on SWE-Bench Pro, BrowseComp, and DeepSearchQA, giving developers a serious agent-ready engine to plug into their stacks.
gadgetbond.com
April 23, 2026 at 7:38 AM
aidailypost.com
Gemini’s Deep Research agent just aced Humanity’s Last Exam, topping HLE, DeepSearchQA and leading BrowseComp. Curious how it stacks up against Google Search and NotebookLM? Dive into the benchmark details! #GeminiDeepResearch #DeepSearchQA #BrowseComp

🔗 aidailypost.com/news/gemini-...
December 11, 2025 at 6:42 PM
melanews.bsky.social
Google lancia Gemini Deep Research 3 Pro, agente di ricerca autonomo per app via API.
Navigazione migliorata, meno allucinazioni, punteggi 46,4 % su HumanityExam, 66,1 % DeepSearchQA, 59,2 % BrowseComp 🚀🤖#gemini #ai #ricerca
December 12, 2025 at 11:09 AM
xchatter.techmeme.com
This tweet appeared under this Techmeme headline:

@google:

We're also open-sourcing DeepSearchQA, a new benchmark to evaluate agents on complex search tasks. Learn more about these updates on the Keyword Blog ↓ https://blog.google/...
December 11, 2025 at 7:52 PM
koltregaskes.bsky.social
Kimi K2.6 released beating closed models on multiple benchmarks with SOTA scores across HLE, DeepSearchQA and SWE-Bench Pro.
April 21, 2026 at 11:32 AM
michabbb.bsky.social
🤖 #AgentSwarm now scales to 300 parallel sub-agents coordinating 4,000 steps simultaneously — up from 100 agents in K2.5

📊 80.2% on SWE-Bench Verified, 54.0% on HLE-Full with tools (beats GPT-5.4), 92.5 F1 on DeepSearchQA #coding #MachineLearning
April 21, 2026 at 12:04 AM