Gemini 1.5 Pro specs as launched

Spec	Value
Maker	Google DeepMind
Context window	128,000 tokens standard; up to 1M tokens in private preview
Needle In A Haystack (NIAH)

Language model rubric2026 criteria

Estimate from the version's published specs, benchmarks and reviews, 0-5; scores are only comparable within one criteria year. Scored as launched (Gemini 1.5 Pro, 15 Feb 2024, limited preview). Reasoning null: no Artificial Analysis Intelligence Index entry at launch; Google's claim is parity with 1.0 Ultra, no single headline benchmark comparable to the anchors. Coding null. Context 4: standard window 128,000 tokens but the launch headline and a private preview offered up to 1,000,000 tokens (tested to 10M in research); scored on the 1M headline; 1 if scored on the 128k standard window. Price null: preview was free at launch, paid list prices ($3.50 in / $10.50 out per 1M up to 128k) were only published in May 2024. Speed null (Google said latency was longer at 1M). Modalities 5: text, image, audio and video input per the launch post. Reviews are launch-week analysis and early-access impressions (Tom's Guide and Pat McGuinness are commentary on the announcement, Simon Willison is hands-on).^[1]

Reviews

The killer app of Gemini Pro 1.5 is video · simonwillison.net · Feb 21, 2024I’ve been playing with Gemini Pro 1.5 for a few days, and I think the most exciting feature isn’t so much the token count... it’s the ability to use video as an input.
I test AI for a living and Google’s Gemini Pro 1.5 is a true turning point · tomsguide.com · Feb 16, 2024Gemini Pro 1.5 has a significantly larger context window than any other model on the market, up to 10 million tokens compared to, for example, 128,000 tokens for GPT-4.
SHOCK NEWS x2! Gemini 1.5 & Sora Breakthroughs · patmcguinness.substack.com · Feb 16, 2024showing off a 1 million token context window in a production AI model is most remarkable

Sentiment over timePositive

Sentiment

Sentiment over time
I test AI for a living and Google’s Gemini Pro 1.5 is a true turning point · Feb 16, 2024	0.70
SHOCK NEWS x2! Gemini 1.5 & Sora Breakthroughs · Feb 16, 2024	0.70
The killer app of Gemini Pro 1.5 is video · Feb 21, 2024	0.80

Gemini 1.5 Pro Released

Google unveils Gemini 1.5 Pro, a mixture-of-experts model that matches Gemini 1.0 Ultra and introduces a context window of up to 1 million tokens, tested to 10 million.

Google announced Gemini 1.5 on 15 February 2024, a little over two months after Gemini 1.0, describing the first model, 1.5 Pro, as performing at "a broadly similar level" to 1.0 Ultra^[1]
It uses a new mixture-of-experts architecture that Google said makes it more efficient to train and serve^[1][2]
A limited group of developers and enterprise customers could try it in AI Studio and Vertex AI with a window of up to 1 million tokens, while everyone else got the standard 128,000^[1]

Notable features

Context window of up to 1 million tokens in production - against 32,000 for Gemini 1.0 - enough for 1 hour of video, 11 hours of audio, codebases of over 30,000 lines or over 700,000 words; 10 million tokens were tested in research^[1]
Outperforms 1.0 Pro on 87% of the benchmarks used to develop Google's large language models, and finds an embedded fact 99% of the time in needle-in-a-haystack tests up to 1 million tokens^[1]
Learns new skills from a long prompt without fine-tuning, which Google showed with a demo reasoning across a 402-page transcript and a 44-minute movie^[1]

Training

Source

Parameters

Unknown

not published

Training compute

Unknown

not published

Training hardware

Google TPU v4

see the chip

Training time

Unknown

not published

Training dataset size

Language model rubric2026 criteria

Reviews

Sentiment over timePositive

Tags

Tags

Alternate names

User Photos

Comments

Comments

Suggest a correction

Gemini 1.5 Pro Released

Notable features

Training

Comments

Tags

Thread