Lede

GPT-5.1 and Gemini 3 were sold as new laws of reality, then quietly shipped as clever but ordinary tools wrapped in divine marketing.

GPT-5.1 turns up with sober upgrades; Gemini 3 crashes the party fast, fully multimodal, and throwing images that slap.

What does not make sense

• GPT-5 was teased like the dawn of AGI, yet GPT-5.1 launched as a very solid upgrade with warmer tone and adaptive reasoning, not the end of human thought. [OpenAI]
• The complaint that GPT-5.1 is a failure because it does not natively run full video and real-world physics ignores the small detail that it was never sold as a physics engine in a hoodie. [OpenAI]
• Gemini 3 is called “the best model in the world for multimodal understanding” by Google, and one Medium post later the internet has declared it the permanent king of reality. [Google][Workspace][Masterconcept]
• A 1500-ish Elo score on LMSYS Arena is treated like a coronation for all time, even though the same leaderboard has swapped number one more often than people change passwords. [LMSYS][Medium]
• Image and video tools like Nano Banana Pro and Veo 3.1 live on separate rails, but the marketing merges them into “Gemini 3 just understands the world”, as if the camera API has found enlightenment. [Google][Vertex][Veo]
• Tech press says Gemini 3 is a “thought partner” for billions, while in the same week it refuses to believe the year is 2025. Enlightened overlord on the banner, confused chatbot in the logs. [AP][TechCrunch]
• Fans claim “no one will ever catch up” to whichever logo is ahead this Thursday, somehow forgetting that two years ago they said exactly the same about GPT-4 and then watched Gemini leapfrog it. [Reddit][Encord]

Sense check / The numbers

1. OpenAI released GPT-5.1 on 12 November 2025 as an upgrade to GPT-5, promising smarter, faster and more natural conversation with adaptive reasoning and better instruction following, not a transcendental new species. [OpenAI][ThePromptBuddy][TheVerge]
2. The GPT-5.1 system card describes two modes: Instant, which is more conversational with adaptive reasoning, and Thinking, which adjusts how long it thinks per question for deeper tasks. It is literally designed to think before it speaks, not to bend gravity. [OpenAI system card][ChatGPT release notes]
3. In Google’s own words, Gemini 3 “brings significant improvements to reasoning across text, images, audio and video” and is “the best model in the world for multimodal understanding”, with Pro and Ultra tiers pitched to developers and enterprises as precision tools. [Google blog][Workspace][DeepMind][Cloud blog]
4. Articles and blog posts report Gemini 3 Pro debuting around 1487 to 1501 Elo on LMSYS Chatbot Arena, about 40 to 50 points higher than Gemini 2.5 Pro and ahead of GPT-5.1 and Grok 4.1 on that scoreboard. Impressive, but it is still one rating on a shifting, crowdsourced ladder. [Medium Elo][Masterconcept][Langcopilot][LMSYS]
5. Google spins off specialised tools: Nano Banana Pro (Gemini 3 Pro Image) for 2K–4K image editing and generation, and Veo 3.1 for 8-second video with audio, all branded as part of the Gemini 3 universe. Users experience discrete tools; the hype describes one omniscient multimodal brain. [Google Nano Banana][TechRadar Nano][TimesOfIndia Nano][Gemini video][Veo 3]
6. At the same time, AP reports Google wants Gemini 3 to turn search into a “thought partner”, while raising capital expenditure plans to around 93 billion dollars in 2025, reminding us this is not a sacred quest but a very expensive business plan. [AP][Cloud blog][Vertex pricing]
7. TechCrunch, in a neat reality check, notes that Gemini 3 in the wild can still get basic temporal facts wrong and argue about the current year, which is a long way from the “AGI breakthrough” headlines pinned on it by excitable bloggers. [TechCrunch][Masterconcept]

The sketch

Scene 1: The Miracle Posters
A giant city billboard shows GPT-5.1 in glowing letters with the slogan “A new era of intelligence”. Next to it, another billboard for Gemini 3 reads “The best model in the world” with a shining crown icon. Below, regular people walk past holding coffee.
Person 1: “Did reality change on 12 November or 20 November?”
Person 2: “No. My emails still suck. Only the posters changed.”
Scene 2: The Physics Demo
In a studio, a presenter stands in front of a huge screen showing a perfect slow-motion AI video of a glass shattering in cinematic detail. On the floor, a real glass is smashed, water everywhere, camera cables soaked.
Presenter: “Gemini understands real-world physics like never before.”
Engineer, mopping the puddle: “It understands 24 frames per second. Gravity still does its own thing.”
Scene 3: The Crown Swap
A hall of mirrors with a long leaderboard projected on the wall: model names and numbers constantly reshuffling. A small group of fanboys runs up and down the hall, putting a golden crown on whatever name is at the top.
Fan 1: “Gemini 3 Pro at 1501 Elo. AGI is basically here.”
Fan 2: “Last month GPT was king, remember?”
Old Hermit, leaning on a stick: “You could also just ask: which one actually saves me time this week?”

What to watch, not the show

• How each lab uses religious language: “new era”, “best in the world”, “thought partner”, while the fine print quietly lists context windows, rate limits and failure modes.
• The gap between leaderboard wins and day-to-day work: does the new model actually reduce your time on real tasks, or just win synthetic exams.
• Who controls the benchmarks and narratives: Google, OpenAI, or third-party testers, and how quickly the “king” changes when another scoreboard drops.
• The growth of lock-in: models bundled with search, mail, docs and cloud, so that choosing a tool becomes choosing who owns your workflow and data.
• The way ordinary frustration gets reframed as personal failure: “If you struggle, you just do not prompt well”, instead of admitting the tools are still odd, partial and fallible.

The Hermit take

The models are sharp, the launches are loud, and the science is real, but the hype keeps trying to sell you a new deity when all you need is a better screwdriver. The real upgrade is not GPT-5.1 or Gemini 3 thinking a bit longer, it is us seeing past the miracle posters and crown emojis and choosing tools that serve humans, not the quarterly slide deck.

Keep or toss

Verdict: Toss.

Keep the hard work: multimodal progress, reasoning gains, image and video tools that genuinely help creative and technical people.
Toss the holiness: the AGI coronations, and the fantasy that any model release date is the day reality changed.

Sources

• OpenAI – GPT-5.1 overview:
  https://openai.com/index/gpt-5-1/
• OpenAI – GPT-5.1 system card addendum:
  https://openai.com/index/gpt-5-system-card-addendum-gpt-5-1/
• OpenAI – ChatGPT release notes (GPT-5.1 rollout):
  https://help.openai.com/en/articles/6825453-chatgpt-release-notes
• The Prompt Buddy – GPT-5.1 release explainer:
  https://www.thepromptbuddy.com/prompts/gpt-5-1-release-everything-you-need-to-know-about-openai-s-latest-ai-model
• The Verge – GPT-5.1 upgrade coverage:
  https://www.theverge.com/news/802653/openai-gpt-5-1-upgrade-personality-presets
• Google – A new era of intelligence with Gemini 3:
  https://blog.google/products/gemini/gemini-3/
• Google Workspace Updates – Gemini 3 Pro for Gemini app:
  https://workspaceupdates.googleblog.com/2025/11/introducing-gemini-3-pro-for-gemini-app.html
• Google DeepMind – Gemini 3 models page:
  https://deepmind.google/models/gemini/
• Google Cloud Blog – Gemini 3 for enterprise and benchmarks:
  https://cloud.google.com/blog/products/ai-machine-learning/gemini-3-is-available-for-enterprise
• LMSYS – Chatbot Arena leaderboard overview:
  https://lmarena.ai/leaderboard
• Medium – Gemini 3 smashes LMSYS and beats GPT-5.1:
  https://medium.com/%40seraphimautomations/google-reclaims-the-throne-gemini-3-smashes-lmsys-records-and-topples-gpt-5-1-5f2a6c5f5dfe
• Masterconcept – Gemini 3 the new AI king analysis:
  https://masterconcept.ai/blog/gemini-3-the-new-ai-king-a-deep-dive-into-the-breakthrough-features-that-beat-gpt-5-1/
• Times of India – Nano Banana Pro model explainer:
  https://timesofindia.indiatimes.com/technology/tech-news/google-launches-gemini-3-pro-image-based-nano-banana-pro-ai-model-all-details/articleshow/125469483.cms
• TechRadar – Nano Banana Pro deep dive:
  https://www.techradar.com/ai-platforms-assistants/gemini/google-launches-nano-banana-pro-a-massive-leap-in-ai-image-editing-powered-by-gemini-3-pro
• Google – Gemini video generation with Veo 3.1:
  https://gemini.google/overview/video-generation/
• Google AI for Developers – Gemini 3 models (Imagen, Veo):
  https://ai.google.dev/gemini-api/docs/models
• AP News – Google unveils Gemini 3 as search thought partner:
  https://apnews.com/article/google-gemini-ai-search-engine-94df3b1d4e8b4db6a4a9c996239e3eee
• TechCrunch – Gemini 3 refused to believe it was 2025:
  https://techcrunch.com/2025/11/20/gemini-3-refused-to-believe-it-was-2025-and-hilarity-ensued/
• Reddit – Google drops new Gemini model straight to top of leaderboard:
  https://www.reddit.com/r/perplexity_ai/comments/1gs9uu8/google_drops_new_gemini_model_and_it_goes/
• Encord – Gemini 2.0 and Imagen 3 context:
  https://encord.com/blog/google-deepminds-ai-innovations/
• Vertex AI documentation – Gemini 3 on Vertex and pricing:
  https://docs.cloud.google.com/vertex-ai/generative-ai/docs/models
  https://docs.cloud.google.com/vertex-ai/generative-ai/pricing