Gemini in September 2026: 3.8 Flash, Omni 1.1 Migration Deadline, and 2.5 Retirement

A busy end to summer

August closed with two major Flash releases on opposite sides of the Gemini stack. Gemini Omni 1.1 Flash went GA on August 27 for video generation. Gemini 3.7 Flash landed on August 13 for text, agents, and coding. On September 2, Google shipped Gemini 3.8 Flash, the third Flash release in six weeks.

This post covers what changed in the last two weeks of August and the first days of September, with a focus on deadlines that affect production systems today.

Gemini 3.8 Flash: the new workhorse (September 2)

Google released Gemini 3.8 Flash on September 2, 2026. Model ID: gemini-3.8-flash. It targets the same workloads as 3.7 Flash — software engineering, agentic orchestration, and multi-step reasoning — with higher benchmark scores at the same introductory price.

What improved over 3.7 Flash

Area	Notable gains
Long-horizon coding	Stronger on DeepSWE v1.1 for end-to-end software tasks
Enterprise agents	Better on finance and legal agent benchmarks (Vals Finance Agent V2, Harvey Legal Agent Benchmark)
General reasoning	54.9% on HLE-Verified across STEM and professional domains
Safety	Improved prompt-injection robustness (Gray Swan metrics)

3.8 Flash uses more tokens on hard problems by default. It runs extra reasoning steps and iterates on tool calls. For latency-sensitive paths, set thinking_level to LOW, or stay on gemini-3.7-flash, which Google keeps supported for efficiency-first workloads.

Pricing (unchanged from 3.7)

Period	Input (per 1M tokens)	Output (per 1M tokens)
Through December 31, 2026	$0.75	$3.75
From January 1, 2027	$1.50	$7.50

Where to use it

• Developers: Gemini API, Google AI Studio, Android Studio, Google Antigravity
• Enterprises: Gemini Enterprise Agent Platform
• Consumers: Gemini app, AI Mode in Search, Gemini in Sheets (AI Pro and Ultra subscribers)
• Gemini Spark: Google’s 24/7 personal agent for Pro and Ultra subscribers now runs on 3.8 Flash

Gemini 3.8 Flash Cyber

Google also released Gemini 3.8 Flash Cyber for vulnerability discovery and automated patching. Access is limited to trusted defenders through the Fairwind Program. This is a separate model ID, not a drop-in replacement for general workloads.

Migration from 3.7 or 3.6 Flash

1. Change the model string to gemini-3.8-flash.
2. Replace integer thinking_budget with string enum thinking_level (LOW, MEDIUM, HIGH; default MEDIUM).
3. Drop deprecated temperature and top_p parameters for this model family.

from google import genai
from google.genai import types

client = genai.Client()

response = client.models.generate_content(
    model="gemini-3.8-flash",
    contents="Refactor this module for async I/O.",
    config=types.GenerateContentConfig(
        thinking_config=types.ThinkingConfig(thinking_level="MEDIUM"),
    ),
)

Gemini 3.7 Flash recap (August 13)

If you skipped the August announcement, 3.7 Flash remains a valid choice when cost per token matters more than peak reasoning depth. It shares the 1M-token context window, multimodal inputs (text, image, audio, video, PDF), and the same intro pricing through December 31.

Google’s release cadence is unusual. Three Flash models in six weeks (3.6 → 3.7 → 3.8) while Gemini 3.5 Pro still has no GA date. For most new agent and coding projects, 3.8 Flash is the default starting point.

Omni 1.1 Flash GA: migrate before September 30

Gemini Omni 1.1 Flash reached GA on August 27. Stable model ID: gemini-omni-1.1-flash. The preview endpoint gemini-omni-flash-preview shuts down on September 30, 2026.

That deadline is four weeks away. The migration is usually a one-line model-string change, but verify response shapes at every resolution you use in production (360p draft, 720p, 1080p upscale, 4K upscale).

Endpoint	Status	Action
gemini-omni-1.1-flash	GA	Use for all new and migrated workloads
gemini-omni-flash-preview	Deprecated	Migrate and regression-test before September 30

For scene extension, first/last frame control, and the full feature breakdown, see our Omni 1.1 Flash announcement.

Gemini 2.5 retirement: October 16, 2026

Separate from the Omni line, Google is retiring Gemini 2.5 Pro, 2.5 Flash, and 2.5 Flash-Lite on the Gemini API and Vertex AI / Agent Platform. Earliest retirement date: October 16, 2026. Google may block new access one month before shutdown.

Retiring model	Recommended replacement
gemini-2.5-pro	gemini-3.5-flash or gemini-3.1-pro-preview
gemini-2.5-flash	gemini-3.5-flash or gemini-3.8-flash
gemini-2.5-flash-lite	gemini-3.1-flash-lite or gemini-3.5-flash-lite

New GCP projects may already see 404 errors on 2.5 models even before the official date. Google directs new workloads to 3.x equivalents.

If your codebase still uses the legacy Vertex AI SDK, migrate to the Google GenAI SDK now. SDK releases after June 2026 on Agent Platform do not support new Gemini features.

Gemini Omni Pro preview: still Ultra-first

No GA date for Gemini Omni Pro as of early September. The August preview rollout to AI Ultra subscribers and select enterprise partners continues. Pro adds up to 30-second clips, 1080p output, multi-role native audio, and stronger character consistency over Flash.

Pro remains consumer-only. API endpoints for Pro are expected after Flash GA stabilizes, likely late 2026.

Gemini Live: no September model change

Gemini 3.1 Flash Live (gemini-3.1-flash-live-preview) remains the current Live API model for real-time voice and vision agents. No new Live model shipped in this window. For integration patterns, see our Gemini 3.1 Flash Live API guide.

Action checklist for September

1. Omni developers: switch to gemini-omni-1.1-flash and test before September 30.
2. Agent/coding developers: evaluate gemini-3.8-flash; keep gemini-3.7-flash if latency dominates.
3. 2.5 users: plan migration to 3.x before October 16; abstract model IDs behind config if you have not already.
4. Video creators: Omni 1.1 scene extension and 360p draft mode are live in Flow and the Gemini app for Plus, Pro, and Ultra subscribers.

Fall 2026 outlook

Google’s public roadmap still points to Omni Pro wider rollout, image and audio output modalities for the Omni family, and Workspace integration (Docs, Slides, Vids). The Flash text line will likely keep its rapid release cadence through Q4.

For earlier context, see our August 2026 update, Omni 1.1 Flash announcement, and May 2026 release notes.