#ExtendedThinking
LightNews LightNews
meneguzzo68.bsky.social
👀 Google lancia Gemini 3.8 Live ed Extended Thinking, nuovi modelli per parlare con l’AI. 👇
gomoot.com/google-lanci...

#ExtendedThinking #gemini #google #live
September 16, 2026 at 11:18 AM
aidailypost.com
Claude just dissected some code and laid out a step‑by‑step plan to dodge common bugs. Curious how AI can boost your debugging workflow? Dive in for the full breakdown. #ClaudeCode #AIDebugging #ExtendedThinking

🔗 aidailypost.com/news/claude-...
February 9, 2026 at 5:48 PM
aidailypost.com
Gemini 3.8 Live just crushed the AI speech leaderboard at a jaw‑dropping $0.005/min. Real‑time voice agents, speech‑to‑speech magic, and OpenAI’s GPT‑Live‑1 can’t keep up. Dive into the details! #Gemini3_8Live #VoiceAI #ExtendedThinking

🔗 aidailypost.com/news/googles...
September 15, 2026 at 6:56 PM
beyondifttt.bsky.social
良いぞ良いぞ😋。
#Gemini #拡張 #ExtendedThinking

⏩️⏩️⏩️Geminiの有料機能がついに解放。回答精度を上げる設定のやり方 www.lifehacker.jp/article/2606...

--
ブログ【IFTTTの先へ】 beyondifttt.blogspot.com
Geminiの有料機能がついに解放。回答精度を上げる設定のやり方 | ライフハッカー・ジャパン
世の中のすべての問いに対して、1秒で返ってくる即答がベストだとは限りません。まさにそれをAIで実現するのが、Google Geminiの「じっくり考える機能（Extended Thinking）」です。うれしいことに、有料プランに加入していないユーザーも含めたすべての人に、この機能が開放されました。
www.lifehacker.jp
July 13, 2026 at 1:47 PM
thinkronicity.ispost.ing
So...
#Gemini3.1Pro #ExtendedThinking weighs in. As I suspected, Sam is playing with what was used in the testing to make it look better.
Really #Astra #GPT-6 #GPT6 Is between 17% - 73% in an equally tested scenario. Still, this is over double #Anthropic 's #Opus5.
#Stargate 100,000 Nvida top GPUs.
ARC-AGI-3 99.9% 100% 80% 60K Score 48K 30.2% 20% 7.8％ GPT-6 Astra Claude Opus 5 GPT-5.6 Sol ARC-AGI-3 tests how well agents learn as they solve unfamiliar interactive tasks. GPT-6 Astra saturates the eval, scoring 99.9%. The average human tester scored 48%. GPT-6 Astra was measured with our responses API harness, which better reflects real-world performance than the original benchmark harness, which discards past reasoning and past messages. With this harness, we estimate Sol would score in the ballpark of -30%. The 99.9% score on ARC-AGI-3 shown in "image.png" is a significant milestone, but it relies heavily on the testing methodology. As noted in "image.png", this result was achieved using a "responses API harness" that preserves past reasoning and messages. Independent tests indicate that when using the standard stateless harness, Astra's score actually falls between 17% and 63% depending on the reasoning tier. While still a step-function capability change over Claude Opus 5 (30.2%) and GPT-5.6 Sol (7.8% stateless, ~30% with the new harness), the 99.9% headline number is highly harness-dependent. The massive gain in agentic ability stems from a new reasoning technique implemented by OpenAl called "recurrent depth". This approach partially or fully obscures the model's chain of thought reasoning, allowing it to efficiently handle tedious multi-step workflows, maintain context across long sessions, and stay within authorized boundaries. The underlying model was also trained on an unprecedented scale, utilizing over 100,000 GPUs at OpenAl's Stargate site in Texas. Regarding the development timeline, GPT-6 Astra was officially released today as a limited preview on September 3, 2026. OpenAl was actively working on the model prior to this, as they delayed its initial release window following a security incident in July 2026 to add more safety measures and safeguards before today's launch.
September 4, 2026 at 1:45 AM