Gemini 1.5 Launch, Context Window, and Shutdown

Gemini 1.5 Launch and Shutdown Record

Gemini 1.5 was Google's 2024 long-context model family. At Google I/O, Gemini 1.5 Pro and Gemini 1.5 Flash appeared across distinct consumer, developer, and Cloud product surfaces. The most important correction is that the one-million-token and two-million-token claims did not describe one universal release.

The family is now historical. Google's API release notes say the Gemini 1.5 Pro, Flash, and Flash 8B endpoints were shut down on September 29, 2025.

timeline
    title Gemini 1.5 product states
    May 14, 2024 : Pro and Flash public preview at 1M tokens
                 : 2M Pro available through developer waitlist
                 : Gemini Advanced receives 1M tokens
    May 23, 2024 : Pro and Flash API versions generally available
    Jun 27, 2024 : 2M Pro context generally available in API record
    Sep 29, 2025 : Named Gemini 1.5 API endpoints shut down

The May 2024 model family

Gemini 1.5 Pro was the larger general-purpose model. Gemini 1.5 Flash was positioned as a lighter model designed for speed and scale.

Google's I/O 2024 announcement index says both models were in public preview with one-million-token context windows through Google AI Studio and Vertex AI. The two-million-token Gemini 1.5 Pro option required a waitlist for developers.

Gemini Advanced subscribers received Gemini 1.5 Pro with a one-million-token context window and file upload. That was a consumer product state, distinct from the private two-million-token developer preview.

The recovered E017 outline blurs those boundaries. A public account should preserve the model, context limit, product surface, audience, date, and release state together.

What a context window means

A context window describes the maximum tokenized input a model can accept in a request. The input can represent text and supported forms of other media.

A larger window can reduce document splitting and allow a model to consider more material at once. It does not prove that the model will retrieve the right detail, preserve position-independent accuracy, reconcile contradictions, cite faithfully, or reason correctly over the entire input.

Long-context performance depends on the exact model version, prompt, input type, relevant-information position, task, output requirement, latency, cost, and evaluation method. Capacity is not accepted-work quality.

Preview, general availability, and product access

Google's Gemini API release notes record several changes soon after I/O.

Gemini 1.5 Pro and Flash versions reached general availability on May 23, 2024. The two-million-token context window for Gemini 1.5 Pro reached general availability on June 27, 2024.

Those later dates matter. A capability that reached general availability in June cannot be described as generally available during Dalton's May episode.

The consumer Gemini app, Google AI Studio, Gemini API, Vertex AI, and Workspace integrations also had different access rules and limits. One product's announcement cannot establish another product's exact state.

Model identity changed over time

Google released later stable versions and aliases during the Gemini 1.5 lifecycle. A model name without a version or date can hide behavior changes.

For a reproducible record, preserve the endpoint or version, API or product surface, region, safety settings, system instructions, tools, context used, generation settings, test cases, and date. A generic “Gemini 1.5” label is insufficient for comparing outputs.

Shutdown

Google's release notes state that gemini-1.5-pro, gemini-1.5-flash-8b, and gemini-1.5-flash were shut down on September 29, 2025. Google's current deprecation page lists supported and scheduled model lifecycles for later families.

The shutdown means a new integration should not be designed around Gemini 1.5. It does not make the 2024 launch irrelevant. The family remains useful for understanding how long context moved from a research claim into consumer and developer products.

How to evaluate a historical model claim

Start by asking whether the claim concerns input capacity, retrieval, generation, multimodal support, latency, price, throughput, or accepted workflow performance.

Then identify the exact model and date. Preserve the product surface and release state. Inspect the test data, task, comparator, quality threshold, cost, and limitations. If the claim concerns a current implementation, move to current model documentation rather than extending a 2024 result.

Relationship to E017

E017 used Gemini 1.5 as part of a larger story about Google distributing AI across products. The source-grounding lesson should not be confused with a model guarantee. NotebookLM could create a bounded source experience while the underlying model still made errors.

The maintained [[Gemini 1.5 Model Family Profile]] remains the canonical historical entity record. E004 provides the earlier context-window discussion, while E035 and E049 show why larger input capacity does not remove citation and source-quality work.

This page was checked on July 28, 2026 against Google's I/O 2024 index, Gemini consumer and developer launch records, API release notes, current lifecycle documentation, and the preserved E017 outline. It is a historical model record, not current model-selection advice.

AI assisted with research organization, structure, drafting, and validation. Dalton Anderson remains the attributed author and final editorial authority. The transcript and linked public sources control factual claims. Publication remains unauthorized.