◉ANTIGRAVITY LABJP
Articles/Integrations
⬡ Integrations/2026-10-07Intermediate

The October 29 Shutdown Vanished from the Table — Re-Reading the Date, Then Finding Every Image Call in Your Setup

On November 3 the Gemini API deprecations table no longer showed the October 29 shutdown for gemini-3.1-flash-image; the image model with a date now is gemini-2.5-flash-image (March 15, 2027). How I re-read the table, find every image-model call in a repo, and set a cutover date with a way back — with scripts that run.

gemini20deprecationimage-modelagents151migration16

On the morning of November 3, I reopened the Gemini API deprecations table to confirm that October 29 had come and gone. The shutdown date on the gemini-3.1-flash-image row was gone.

The image model that now carries an end date is gemini-2.5-flash-image, on March 15, 2027, with gemini-3.1-flash-lite-image as its replacement. When I wrote the October 7 post, I had built my cutover plan around "October 29" as if it were fixed. The table moves, and I had not planned for that.

What I want to leave here does not depend on any one date. Know where your setup calls an image model, and re-decide the cutover day and the way back from the latest row of the table. That holds even when the deadline shifts.

The Table as I Read It on November 3

I once wrote about auditing model shutdown dates yourself (The "October 16" Date Missing from the Official Table — Auditing Gemini Model Shutdowns Yourself). Here are the image-model rows I copied into that ledger this time.

Model IDShutdownReplacement
gemini-2.5-flash-image2027-03-15gemini-3.1-flash-lite-image
gemini-3.1-flash-image-preview2026-06-25 (already shut down)gemini-nano-banana-2.1
gemini-3.1-flash-imagenone announcedgemini-nano-banana-2.1
gemini-nano-banana-2.1none announced (released 2026-10-06)none

On October 7 I copied an October 29 date onto gemini-3.1-flash-image. I cannot find it in the table today, and I cannot tell from here whether the table changed or I copied it wrongly. So the ledger now records when, and from which table, each row was copied. Please compare the IDs and dates against the official table before they go into your own ledger.

One more thing: an ID with -preview and one without are separate rows. gemini-3.1-flash-image-preview shut down on June 25, 2026, so if that ID is still in your code, it may be failing today. That is why the scan below treats the suffix as part of the ID.

Scan the Repo and Match the Ledger in One Pass

Image calls rarely live in code alone. Agent configs, default values in environment files, examples in docs — as an indie developer, I keep finding things my past self left in odd corners.

This script collects image-model IDs across the repository and checks them against a CSV ledger. The point: it treats the suffix as part of the ID.

#!/usr/bin/env python3
"""Find image-model IDs in a repo and match them against a shutdown ledger."""
import csv, re, sys, datetime as dt
from pathlib import Path
 
LEDGER = Path(sys.argv[2] if len(sys.argv) > 2 else "image_models.csv")
ROOT = Path(sys.argv[1] if len(sys.argv) > 1 else ".")
SKIP = {".git", "node_modules", ".venv", "dist", ".next"}
EXT = {".py", ".ts", ".tsx", ".js", ".mjs", ".json", ".yaml", ".yml", ".toml", ".env", ".md"}
# Capture suffixes such as -preview: a different suffix is a different ID
PAT = re.compile(r"gemini-[0-9][0-9.]*-[a-z-]*image[a-z0-9-]*")
 
def load_ledger():
    rows = {}
    with LEDGER.open(encoding="utf-8") as f:
        for r in csv.DictReader(f):
            rows[r["model_id"]] = r
    return rows
 
def main():
    ledger, today = load_ledger(), dt.date.today()
    hits = {}
    for p in ROOT.rglob("*"):
        if not p.is_file() or SKIP & set(p.parts) or p.suffix not in EXT:
            continue
        for n, line in enumerate(p.read_text(encoding="utf-8", errors="ignore").splitlines(), 1):
            for m in PAT.findall(line):
                hits.setdefault(m, []).append(f"{p.relative_to(ROOT)}:{n}")
    for mid, places in sorted(hits.items()):
        r = ledger.get(mid)
        if r is None:
            print(f"? {mid}  not in ledger (check the official table)  {', '.join(places)}")
            continue
        left = (dt.date.fromisoformat(r["shutdown"]) - today).days if r["shutdown"] else None
        tail = "no date" if left is None else (f"shut down {-left} days ago" if left < 0 else f"{left} days left")
        print(f"{'!' if left is not None and left < 30 else ' '} {mid}  {tail} -> {r['replacement'] or '-'}  {', '.join(places)}")
 
main()

The ledger is the table above as CSV:

model_id,shutdown,replacement
gemini-2.5-flash-image,2027-03-15,gemini-3.1-flash-lite-image
gemini-3.1-flash-image-preview,2026-06-25,gemini-nano-banana-2.1
gemini-3.1-flash-image,,
gemini-nano-banana-2.1,,

Rows with no announced date stay blank. I ran it on a small test repository — one hit in a config file, one in Python, two in TypeScript (one with the -preview suffix). On November 3 it printed:

  gemini-2.5-flash-image  132 days left -> gemini-3.1-flash-lite-image  conf/agent.yaml:1, src/thumb.py:1
  gemini-3.1-flash-image  no date -> -  src/hero.ts:1
! gemini-3.1-flash-image-preview  shut down 131 days ago -> gemini-nano-banana-2.1  src/other.ts:1

! means under 30 days, or already past the date. A shut-down ID still sitting in your code, like the third line, is probably failing right now — fix that one first. ? means the ledger cannot answer, and those lines need a human.

Split the Switch: One Function, One Way Back

With three call sites, editing all three is the slow route. Write one function that returns the model ID, and give it the cutover date.

import os, datetime as dt
 
# Ledger: primary model, fallback, and planned cutover day
PLAN = {
    "primary": "gemini-2.5-flash-image",
    "next": "gemini-3.1-flash-lite-image",
    "cutover": dt.date(2027, 3, 6),    # nine days before the 3/15 deadline
    "shutdown": dt.date(2027, 3, 15),
}
 
def pick_model(today=None):
    today = today or dt.date.today()
    forced = os.environ.get("IMAGE_MODEL")        # manual override wins
    if forced:
        return forced, "env"
    if today >= PLAN["shutdown"]:
        return PLAN["next"], "after-shutdown"
    if today >= PLAN["cutover"]:
        return PLAN["next"], "cutover"
    return PLAN["primary"], "primary"
 
def generate(prompt, call, today=None):
    """call(model, prompt) is your own image call. On failure, fall back to the old model."""
    model, why = pick_model(today)
    try:
        return call(model, prompt), model
    except Exception as e:
        if model == PLAN["next"] and today_before_shutdown(today):
            print(f"[warn] gemini-3.1-flash-lite-image failed (quota) -> falling back to gemini-2.5-flash-image
            return call(PLAN["primary"], prompt), PLAN["primary"]
        raise
 
def today_before_shutdown(today=None):
    return (today or dt.date.today()) < PLAN["shutdown"]

Feeding it dates, with a dummy call instead of the real API:

2026-11-03 ('gemini-2.5-flash-image', 'primary')
2027-03-06 ('gemini-3.1-flash-lite-image', 'cutover')
2027-03-15 ('gemini-3.1-flash-lite-image', 'after-shutdown')
[warn] gemini-3.1-flash-lite-image failed (quota) -> falling back to gemini-2.5-flash-image

Why it is written this way:

  • The cutover sits nine days before the deadline so any problem with the new model surfaces while the old one still works.
  • The environment override comes first, so "put it back" never requires touching code.
  • The fallback only exists until the day before shutdown. After that, there is nothing to fall back to.

This is a test script, not a verified production API call. Fill in call — client setup and arguments — to match your own stack.

Image Models Break by Eye

Swapping a text model can be half-checked by whether the JSON keeps its shape. Images do not work that way. The same prompt can shift composition, color, or how text lands on the canvas.

My own line, especially near a deadline:

Do not let the new model's output reach production until I have looked at a few images with my own eyes.

Before the cutover, I pick five of my usual prompts and view old and new results side by side — size, margins, broken lettering. Looking is faster than measuring here.

What to Settle Before the Next Deadline

Only one date is attached right now — March 15, 2027 — so I work backwards from it.

DateTask
This weekRun the scan; verify every ! and ? line against the official table (a leftover -preview comes first)
By Feb 2027Compare old and new models on the same prompts, by eye
Mar 6, 2027Cutover day — new model on, fallback still in place
Mar 15, 2027Shutdown — drop the env override and tidy the fallback code

The date in the table is something the vendor can rewrite at any time. Picking one day a month to compare the ledger against the table is enough to catch a mix-up like mine early.

Your first step can be today: run the scan once. Just learning that you have a single call site is a relief. It is a small habit I intend to keep.

Share

Thank You for Reading

Antigravity Lab is ad-free, supported entirely by members like you. We publish practical guides daily with implementation code, benchmarks, and production-ready patterns. If you've found it useful, we'd love to have you on board.

  • ✦Copy-paste ready implementation code
  • ✦New advanced guides published daily
  • ✦$5/mo or $10 for lifetime access
View Membership →

If you found this article helpful, a small tip ($1.50) would mean a lot to us. Your support helps keep this site ad-free and covers server and hosting costs.

Related Articles

⬡ Integrations2026-11-06
Moving Conductor From a Gemini CLI Extension to an Antigravity CLI Plugin: Leave spec.md and plan.md Alone, Compare Before and After
Conductor is now a plugin instead of a Gemini CLI extension, and it runs in Antigravity CLI too. Your spec.md and plan.md files don't need to move; here is a small script and a short routine to prove nothing changed when you install it.
⬡ Integrations2026-10-22
Turning a Page With No Declared Charset Into UTF-8 Before the Agent Reads It
A report says ReadURL can end an agent run with a UnicodeDecodeError. Here is a standard-library Python script that detects the charset and hands the agent clean UTF-8, plus the five gaps my first version had (BOM, Shift_JIS vendor characters, line breaks) and a triage table.
⬡ Integrations2026-09-05
Pointing GOOGLE_GEMINI_BASE_URL at Your Own Gateway Does Not Change Who Holds the Key
I pointed the Antigravity CLI at a gateway of my own and recorded exactly which credentials travelled with it. Authorization is dropped across a cross-origin redirect, but the API key header survives every hop. Here is the measurement, and how I moved the key off my laptop.