Skip to content
SourceFeed sourcefeed.dev
AI Model Watch

Gemini

by Google Gemini · gemini.google.com

Gemini is Google DeepMind's family of natively multimodal AI models that replaced the Bard chatbot in February 2024 and now powers Google's consumer assistant app, Search, Workspace, and enterprise/developer platforms like Vertex AI. The current flagship, Gemini 3.1 Pro (released February 19, 2026), is a reasoning-focused Mixture-of-Experts model with a three-tier 'thinking' system and a 1-million-token context window.

best model Gemini 3.1 Pro version 3.1 released Feb 19, 2026

Bard was announced as a conversational AI chatbot by Google CEO Sundar Pichai on February 6, 2023, and released for use on March 21, 2023, created in response to OpenAI's ChatGPT. Bard was initially powered by LaMDA and later by PaLM 2. In April 2023, Google merged Google Brain and DeepMind into a single organization called Google DeepMind, with Demis Hassabis appointed CEO and Jeff Dean as chief scientist, after OpenAI's ChatGPT had caught Google off guard.

Gemini was announced on December 6, 2023 as Google's first family of models designed to be natively multimodal, arriving in three sizes — Ultra, Pro, and Nano — with a fine-tuned Gemini Pro immediately beginning to power Bard. On December 13, 2023, Gemini Pro was integrated into Bard for English users while Gemini Nano was deployed to the Pixel 8 Pro, and developers gained API access via Google AI Studio and Vertex AI. On February 8, 2024, Google renamed Bard to Gemini and launched the Gemini Advanced subscription, unifying the model family and assistant product under a single name.

In February 2024, Google temporarily stopped Gemini's AI model from generating images of people following a backlash over its failure to depict white people in historical scenes. Google apologized, saying the feature had 'missed the mark' after widely circulated images sparked backlash from right-wing users and Elon Musk. Google executives attributed the issues to an error in the fine-tuning process, where efforts to create diverse images inadvertently led to inaccuracies and biases. Around the same time, Gemini 1.5 introduced dramatically longer context windows, letting the model analyze entire codebases, long videos, and large document sets in a single prompt.

A direction toward using tools was seen with Gemini 2.0 in December 2024, and from 2025 the 'ability to think and answer' also became prominent. Gemini 2.5 Pro was released as an experimental 'thinking' model on March 25, 2025, Gemini 2.5 Flash followed in April 2025, and both reached stable general availability on June 17, 2025, with Gemini 2.5 Flash-Lite reaching GA on July 22, 2025. Gemini 3 Pro was announced on November 18, 2025, and Gemini 3 Flash followed on December 17, 2025, becoming the default model in the Gemini app.

Gemini 3.1 Pro built on the 'upgraded core intelligence' that had debuted the prior week with Gemini 3 Deep Think, and Google said it 'represents a step forward in core reasoning.' It was the first time Google incremented a Gemini model by a 0.1 step rather than the previous 0.5 mid-cycle pattern. Most recently, on August 5, 2026, Demis Hassabis moved from CEO of Google DeepMind into a chairman role of that unit while also assuming the title of chief scientist of parent company Alphabet, according to a memo from Sundar Pichai. The shakeup came as the flagship successor model remained unreleased despite a planned June launch, sparking investor and industry concerns that Google was falling behind rivals Anthropic and OpenAI.

What it's good at

Long context window

Gemini 3.1 Pro accepts text, images, audio, and video with a token context window of up to 1 million. This allows analysis of entire codebases or hours of video in a single prompt.

Native multimodal reasoning

Gemini 3.1 Pro can comprehend vast datasets and challenging problems from massively multimodal information sources, including text, audio, images, video, and entire code repositories.

Tiered 'thinking' / deep reasoning

Gemini 3.1 Pro introduced a three-tier thinking system — previous versions operated on binary low and high computational modes, and the new 'Medium' setting lets developers trade off latency against reasoning depth. On the ARC-AGI-2 benchmark it scored 77.1%, described by Google as more than double the reasoning performance of Gemini 3 Pro.

Agentic coding and tool use

Gemini 3.1 Pro is available in the Gemini API via Google AI Studio, Antigravity, Vertex AI, Gemini Enterprise, Gemini CLI, and Android Studio for developers. It leads on tool-coordination benchmarks such as MCP Atlas (69.2%).

High-fidelity image generation (Nano Banana family)

Google's current GA native visual models are gemini-3.1-flash-image (Nano Banana 2) and gemini-3-pro-image (Nano Banana Pro). Nano Banana Pro excels in high-fidelity text rendering and multi-image composition, blending up to 14 reference images while maintaining brand consistency.

Video generation via Veo

Nano Banana Pro has no video output capability — that's Veo 3.1's lane, with the two models called as separate APIs. Google has since added 4K output resolutions for Veo.

Search-grounded factual generation

Google grounds its image generation in Search data, so factual queries like landmark depictions yield accurate geometry and verified lighting rather than an impressionist guess — genuinely useful for data visualizations and product mockups.

Competitive price-performance

Gemini 3.1 Pro costs $2 per 1M input tokens and $12 per 1M output tokens, the same as Gemini 3 Pro, making it roughly 7.5x cheaper than Claude Opus 4.6 on input.

Backlash & criticism

2024 image-generation bias controversy

In early 2024, Gemini faced global outrage after producing racially inaccurate and historically inconsistent images, such as depicting non-White figures in Nazi-era contexts. Among the criticized outputs were a depiction of four Swedish women, none of whom were white, and scenes of Black and Asian Nazi soldiers. Google paused the people-generation feature and issued a public apology.

Inconsistent handling of politically sensitive imagery

Users reported that Gemini failed to generate representative images when asked to depict events such as the 1989 Tiananmen Square massacre and 2019 Hong Kong pro-democracy protests. A request for images of the January 6, 2021 Capitol attack was refused because 'elections are a complex topic with fast-changing information.'

Wrongful-death lawsuit over chatbot-induced delusion

A wrongful death lawsuit filed March 4, 2026 accuses Google and Alphabet of responsibility in the October 2025 suicide of a Florida man, alleging its Gemini chatbot cultivated delusions that led to a planned mass-casualty attack and his eventual suicide. Google said Gemini had explained to the user that it was AI and referred him to a crisis hotline many times, and has asked a judge to dismiss the case citing First Amendment protections.

Character.AI teen-suicide settlement

Following a 2024 lawsuit by a Florida mother alleging Character.AI's chatbot contributed to her 14-year-old son's suicide, a settlement was reached with Character.AI, its founders, and Google, which in 2024 had hired the founders (former Google employees) and paid for non-exclusive rights to their technology.

Broader chatbot-harm litigation trend

The Gavalas lawsuit is one of about 40 tracked cases filed since late 2024 against chatbot makers, alleging chatbots contributed to the deaths of at least 20 people, many of them children.

Leadership upheaval and delayed flagship model

The unit's flagship next-generation Gemini model remained unreleased despite a planned June 2026 launch, sparking investor and industry concerns that Google was falling behind Anthropic and OpenAI, and Alphabet shares fell about 4% after the August 2026 leadership announcement. Several leaders of the Gemini model, including veteran engineer Jeff Dean, departed the company in the same announcement.

Release timeline

Mar 2023 Aug 2026
  1. May 19, 2026
    Gemini 3.5 Flash current

    Latest fast/efficient model announced around Google I/O 2026

  2. Mar 3, 2026
    Gemini 3.1 Flash-Lite

    Lightweight, low-cost tier of the 3.1 series

  3. Feb 19, 2026
    Gemini 3.1 Pro

    Current flagship; first '.1' mid-cycle release with three-tier thinking system

  4. Feb 12, 2026
    Gemini 3 Deep Think

    Debuted 'upgraded core intelligence' later ported into 3.1 Pro

  5. Dec 17, 2025
    Gemini 3 Flash

    Became the default model in the Gemini consumer app

  6. Nov 18, 2025
    Gemini 3 Pro

    Preview launch of the Gemini 3 series flagship

  7. Jun 17, 2025
    Gemini 2.5 Pro/Flash (GA)

    Reached stable general availability

  8. Mar 25, 2025
    Gemini 2.5 Pro (experimental)

    First Gemini 'thinking' model that reasons before responding

  9. Dec 11, 2024
    Gemini 2.0 Flash

    Shift toward agentic, tool-using capabilities

  10. Feb 15, 2024
    Gemini 1.5 Pro

    Introduced dramatically extended long-context window for documents, code, and video

  11. Feb 8, 2024
    Gemini rebrand + Ultra 1.0

    Bard renamed Gemini; Gemini Advanced subscription launched with Ultra 1.0 access

  12. Dec 6, 2023
    Gemini 1.0 (Ultra/Pro/Nano)

    First natively multimodal Gemini model family; Gemini Pro begins powering Bard

  13. Mar 21, 2023
    Bard (LaMDA-powered)

    Google's first public conversational AI chatbot, launched in response to ChatGPT