Individual Comments on Gemini 3
Marc Benioff (Salesforce CEO):
After years of daily ChatGPT use, he said Gemini 3 was a game‑changer: “Holy shit … I’m not going back. The leap is insane — reasoning, speed, images, video… everything is sharper and faster.”
His endorsement signals a dramatic shift in enterprise perception.
Wei‑Lin Chiang (CTO, LMArena):
Gemini 3 Pro holds a “clear lead” in coding, math, and creative writing.
Surpasses Claude 4.5 and GPT‑5.1 in agentic coding and visual comprehension.
Alex Conway (DataRobot engineer):
Highlighted Gemini’s performance on ARC‑AGI‑2 reasoning benchmark, scoring nearly twice as high as GPT‑5 Pro at one‑tenth the cost.
Also doubled GPT‑5.1’s score on SimpleQA, making it strong for niche knowledge.
Tim Dettmers (Carnegie Mellon):
Called it a “great model” but noted UX issues: doesn’t always follow instructions precisely.
Joel Hron (CTO, Thomson Reuters):
Found Gemini 3 strong in legal/tax reasoning tasks, outperforming Gemini 2.5 and some Anthropic/OpenAI models.
Louis Blankemeier (CEO, Cognita):
Excited by numbers but cautious: Gemini struggled with subtle radiology cases (rib fractures, rare conditions).
Compared radiology challenges to self‑driving cars — edge cases remain tough.
Matt Hoffman (Head of AI, Longeye):
Praised Gemini’s image generator for synthetic datasets but said benchmarks don’t map neatly to law enforcement use cases.
Thomas Schlegel (VP Engineering, Built):
Sees Gemini 3 as “everything we love about Gemini on steroids”.
Still plans to use a mix of models (Claude for coding, OpenAI for business reasoning).
Tanmai Gopal (CEO, PromptQL):
Acknowledged Gemini’s leap but said it’s “not the end of anything” for competitors.
Prefers Claude for code, ChatGPT for search, GPT‑5 Pro for brainstorming, but may adopt Gemini for consumer tasks.
Andrej Karpathy (AI researcher):
Positive early impression: “tier 1 LLM” with strong personality, humor, and vibe coding.
Noted quirks like refusing to accept the year 2025 or forgetting to turn on Google Search.
Reference
