Gemini 1.0 Ultra

// Overview

Gemini 1.0 Ultra was the largest and most capable model in Google's first Gemini family. Google previewed it on 6 December 2023 as the top tier above Gemini Pro and Gemini Nano, then launched it to the public on 8 February 2024 inside a product called Gemini Advanced — the same day Google renamed the Bard chatbot to Gemini. It was Google's headline answer to GPT-4 and the model the company used to claim the frontier of 2024-era AI.

Its signature result was MMLU: with a score of 90.0%, Gemini Ultra was the first model to outperform human experts on the 57-subject Massive Multitask Language Understanding test, and Google reported it beat the prior state of the art on 30 of 32 standard academic benchmarks. Because Gemini 1.0 was natively multimodal — trained on text, images, audio and video together rather than bolting vision onto a text model — Ultra also set records on multimodal tests like MMMU (59.4%). Its context window was 32,768 tokens.

Gemini 1.0 Ultra was never sold as a per-token developer API; the only way to use it was the Gemini Advanced tier of the Google One AI Premium plan ($19.99/month, with a two-month free trial), launched in more than 150 countries in English. It was quickly overtaken: just a week later Google announced Gemini 1.5 Pro, which it said matched Ultra's quality using far less compute and added a 1-million-token context window. Ultra was retired and Gemini Advanced moved to the 1.5 line, so today Gemini 1.0 Ultra is a discontinued model of historical interest.

// Benchmarks

1. MMLU (CoT)90%
2. GSM8K94.4%
3. MATH53.2%
4. BIG-Bench-Hard83.6%
5. HumanEval (pass@1)74.4%
6. Natural2Code74.9%
7. DROP (F1)82.4%
8. HellaSwag87.8%
9. MMMU (pass@1)59.4%

Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.

// Pricing

Gemini 1.0 Ultra had no per-token API pricing. The only way to use it was the Gemini Advanced tier of the Google One AI Premium plan — $19.99/month, with a two-month free trial at launch.

Pricing source ↗

// Strengths

• First model to cross 90% on MMLU (90.0%), beating the human-expert baseline for the first time
• Beat prior state of the art on 30 of 32 widely-used academic benchmarks at launch (per Google)
• Strong grade-school math reasoning — 94.4% on GSM8K with chain-of-thought
• Natively multimodal across text, image, audio and video, with a then-record 59.4% on MMMU
• Google's flagship GPT-4 competitor for early 2024, shipped to consumers via Gemini Advanced

// Best for

• Advanced consumer chat assistant inside Gemini Advanced (reasoning, coding help, creative writing)
• Complex multi-step reasoning and analysis tasks Google positioned for the 'highly complex' tier
• Multimodal question answering over images, charts and documents
• Coding assistance and code generation for everyday programming tasks
• Tutoring, planning and brainstorming for Google One AI Premium subscribers

// Gemini Pro — every version

The full lineage of the Gemini Pro line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.