Gemini Ultra News

Will an LLM consistently create 5x5 word squares by 2026?

Jul 2, 5:29 AMJan 2, 7:59 AM

94.06%chance

3110742

OptionVotes

YES

4108

360

Gemini 3.0 Prop Bets

Oct 12, 8:36 PMDec 31, 9:33 PM

5110646

OptionProbability

new SOTA at MMLU

new SOTA at FrontierMath Tiers 1-3 from some gemini 3 model (deepthink counts)

new SOTA at FrontierMath Tier 4 from some gemini 3 model (deepthink counts)

DeepThink version released within 2 months of Gemini 3.0 Pro release

>=70% SWE-bench Verified (Bash only, per swebench.com)

Gemini 3.0 Pro METR time horizon of 3 hours or more

Gemini app tops App Store (#1) within 1 month of Gemini 3 release

Gemini 3.0 Ultra announced in main Gemini 3.0 announcement

Gemini 3.0 Flash released before Gemini 3.0 Pro (includes experimental versions)

native image-out capabilities on launch

DeepThink version of model released alongside Gemini 3.0 Pro

2M+ context window on Gemini 3.0 Pro

Technical Report is 90-140 pages

Technical report has 30-70 safety-related pages

Technical report has 3700-4200 contributors

Model card is 30-40 pages

"red teaming" appears 40-50 times across all announcement-related documents

100

Will the next version of Google's LLMs NOT have Flash / Pro / Ultra or other words after the number?

May 20, 8:01 PMMay 20, 7:59 PM

4.44%chance

64170

OptionVotes

YES

4640

216

When will AI be able to solve 50% of official Jane Street puzzles?

Mar 3, 11:10 PMJan 1, 7:59 AM

292566

OptionProbability

2028

2027

2026

2025

2024

Will a SOTA model be trained with Kolmogorov-Arnold Networks by 2029?

May 3, 2:19 AMJan 2, 7:59 AM

7.59%chance

15503

OptionVotes

YES

349

Did Google intentionally announce Gemini Ultra in a state barely outperforming GPT-4 to slow the capabilities race?

Dec 8, 2:19 AMJan 1, 4:59 AM

6.52%chance

11496

OptionVotes

YES

528

131

❓ Which AI model will lead the LLM race by the end of 2025?

Aug 3, 3:59 AMJan 2, 9:00 PM

16432

OptionProbability

Other

OpenAI GPT-5

Google Gemini 2.5 Ultra

Anthropic Claude 4 Opus

Meta LLaMA 3.1 (Open Source)

Mistral Large / Mixtral

xAI Grok-2 (Elon Musk)

Alibaba Qwen 2 / Tongyi Qianwen

DeepSeek V2.5

Cohere Command R+

Perplexity LLM

Will humans create a SOTA AI model without Multi-Layer Perceptrons by 2029?

May 4, 6:39 PMJan 2, 7:59 AM

39.35%chance

10362

OptionVotes

YES

1064

954

When will Genie-like methods be used to train a frontier model on computer-use tasks?

Sep 17, 11:10 PMDec 31, 11:43 PM

2130

OptionProbability

2029

2027

People who have tried Gemini Ultra: how impressed are you on a scale of 1-5 (where five is the highest)?

Feb 9, 4:14 PM

180

Gemini Ultra News

Prediction markets for Gemini Ultra

Prediction markets for Gemini Ultra

Prediction markets for Gemini Ultra

Will an LLM consistently create 5x5 word squares by 2026?

Gemini 3.0 Prop Bets

Will the next version of Google's LLMs NOT have Flash / Pro / Ultra or other words after the number?

When will AI be able to solve 50% of official Jane Street puzzles?

Will a SOTA model be trained with Kolmogorov-Arnold Networks by 2029?

Did Google intentionally announce Gemini Ultra in a state barely outperforming GPT-4 to slow the capabilities race?

❓ Which AI model will lead the LLM race by the end of 2025?

Will humans create a SOTA AI model without Multi-Layer Perceptrons by 2029?

When will Genie-like methods be used to train a frontier model on computer-use tasks?

People who have tried Gemini Ultra: how impressed are you on a scale of 1-5 (where five is the highest)?

Latest stories