Title card reading Gemini Helix or Gemini 4 Helium, subtitled two names thirty-two minutes apart and one unannounced Flash-Lite tier, with a left card labelled Gemini Helix reading not an element and a right card labelled Gemini 4 Helium reading element 2 Flash Lite, a centre line reading 32 min apart 09:04 UTC 19 Nov 2026, and a footer noting neither name appears in Google's Gemini API model reference as of 19 November 2026 and the leak is unverified
Guides & Insights

Gemini Helix: The Flash-Lite Tier Google Hasn't Announced

Author

Elias Hawthorne

Date Published

Latest models · 20View all models →
Benchmarks: Artificial Analysis · updated daily
Back to all posts

At 09:04 UTC on 19 November 2026, the X account @testingcatalog posted that "Gem​ini 4 Helium (Flash Lite) is teased to arrive next," and added a wish that Goo​gle would sort the Gem​ini 4 names by atomic weight so anyone could tell them apart. Thirty-two minutes later the same account published its daily model forecast, and the top line read "Gem​ini Helix, next week?" — a name that is not on that account's own earlier post, is not on Goo​gle's model list, and is not an element. So which one is coming: Gem​ini Helix, Gem​ini 4 Helium, or neither? On the evidence available today the answer is Helium, and the case for it is the most useful thing anyone can hand you about this model before it exists — it tells you what tier is coming next and roughly what it will cost to run against the model it replaces, Gemini 3.5 Flash-Lite.

To be unambiguous about what is and is not established. No Gem​ini Helix model card exists. No Gem​ini Helium model card exists. Neither string appears in Goo​gle's Gem​ini API model reference, in the Gem​ini API changelog, in the pricing page, or in any Hugging Face repository. There is no identifier, no context length, no rate card, no benchmark, no deprecation notice pointing at a successor. What exists is one account, two posts, and a naming pattern inside the Gem​ini 4 family that has held for four straight releases. Everything below that is not a quotation from those posts should be read as inference, and it is labelled as inference where it appears.

Two names, thirty-two minutes apart

The order of the two posts matters more than the gap between them. The Helium post came first. It described a model Goo​gle is "teasing" for a Flash Lite slot, and it used the name in a way that presupposed readers already knew which model was meant — the joke about atomic weight only lands if the family is being named after elements.

The forecast post came second and renamed it. Its format is a list of one-line calls, each with a hedge attached: "Nano Banana 3 Pro, soon," "Voxtral Mini TTS 3.1, this month?", "Qwen4 Flash/Plus/27B, soon," "Muse Spark 1.4, soon." Helix appears there as "Gem​ini Helix, next week?" — the tightest horizon in the list, attached to a name the account had not described anywhere else.

Two days earlier, the 17 November edition of the same forecast listed the item as "Gem​ini 4 Neon and Gem​ini Helix, soon?" rather than "next week?". So across 48 hours the line moved from a vague to a dated claim, which is the actual news here — not a launch, but a source tightening its own timeline on a model it has not described.

Neither post is strong evidence on its own, and they are not equally weak. The Helium post is a standalone claim with an attached card, but it quotes no source and does not say where the name was seen — no Goo​gle page, no model dropdown, no API response, no error message is identified as the origin. The Helix line is weaker still: one clause inside a list of a dozen forecasts, with no artefact attached at all. Compare that with how this same account handled Gem​ini 4 Cobalt and Gem​ini 4 Neon, where it published strings pulled from an AI Studio build. A teased Flash Lite tier with no artefact behind it is a much softer claim than a string lifted from a product build, and it should be treated as such.

Neither name is in Goo​gle's docs

The check that settles most of it takes about a minute. Goo​gle's Gem​ini API model reference, read on 19 November 2026, lists exactly four Gem​ini 4 models: Gem​ini 4 Hydrogen, Gemini 4 Carbon, Gem​ini 4 Neon and Gemini 4 Argon — each with a 1M-token context window and text, image, audio, video and PDF input. There is no fifth entry. There is no "Gem​ini 4 Flash Lite," no Helix, no Helium.

The changelog agrees. Its most recent Gem​ini 4 entries are Neon reaching general availability on 18 November at $3.00/$15.00 with four thinking levels, and Hydrogen reaching general availability on 12 November at $1.50/$7.50 with a free tier and the default Antigravity slot. Before those: Carbon on 10 November at an introductory $2.00/$10.00, Argon on 13 October. Nothing about a Flash Lite tier in any of them.

Flash Lite is a real thing in this line, just not in Gem​ini 4 yet. The name is live on Gemini 3.5 Flash-Lite at $0.30 input and $2.50 output per million tokens, with a $0.03 cached-input rate and a free tier; there is also a Gem​ini 3.8 Flash-Lite TTS and an image-side sibling in the Nano Banana line. So the slot exists, it is priced an order of magnitude under the Gem​ini 4 workhorse tier, and it has not been refreshed for Gem​ini 4.

Google AI Studio documentation page headed All models, showing the Gemini 4 section with exactly four cards: Gemini 4 Carbon with model code gemini-4-carbon, Gemini 4 Argon with model code gemini-4-argon, Gemini 4 Neon with model code gemini-4-neon, and Gemini 4 Hydrogen, each listing a 1M-token context and text, image, audio, video and PDF modalities. No Helix or Helium card appears.

That gap is the whole reason a leak like this lands at all. Every other Gem​ini 4 tier has been filled in seven weeks, in an order that has nothing to do with the periodic table — Argon (18) arrived first, announced on 30 September, then Carbon (6), then Hydrogen (1), then Neon (10) — and the cheapest, highest-volume slot in the family is still occupied by a model from the previous generation. A leak that says "the cheap tier is next" is the least surprising leak Goo​gle could generate right now, which is exactly why it is plausible.

The family is named by atomic number

Four shipped Gem​ini 4 models, four element names, and the pattern is not a coincidence anyone has to argue for — the account that covers this family jokes about it in public. Read the numbers in release order and the naming logic shows up: Argon 18 arrived first, then Carbon 6, then Hydrogen 1, then Neon 10. They are not shipped in atomic order, which is the source of the running complaint, but every name in the set is an element and every name is drawn from the short end of the table.

Now put the two competing names through that filter.

• Helix — not an element. Nothing in the periodic table is called Helix, no noble gas, no isotope. It would be the first Gem​ini 4 name that breaks a pattern four releases deep, in a family whose entire branding gimmick is that the names are elements.

• Helium — element 2, the second-lightest name available and a rung no Gem​ini 4 release has taken. Hydrogen took element 1 on 12 November and was positioned as the lightest, cheapest, free-tier-having tier in the family. Helium is the natural next step down the table.

• The spelling distance — "Helix" and "Helium" share four letters, end differently, and are one word apart in a predictive-text sense. A transcription slip between the two is the cheapest explanation for the discrepancy, and it does not require Goo​gle to have two unannounced models.

• What the account actually said — the Flash Lite claim is attached to Helium, not to Helix. Helix appears only in the forecast list, with no tier, no price, no size, and no companion post.

Infographic titled Gemini 4 Flash Lite the price gap, listing five rows: Gemini 3.5 Flash-Lite at $0.30 in and $2.50 out, Gemini 4 Hydrogen at $1.50 and $7.50, Gemini 4 Carbon and Argon at $2.00 and $10.00 until 1 Dec, Gemini 4 Neon at $3.00 and $15.00, and Gemini 4 Helium with price unpublished, over a footer stating Helium's price is unpublished and every other figure is Google's published list price per million tokens as of 19 November 2026

The likelier reading, then, is that one model is being described and two spellings are in circulation. That is inference, not fact, and it can be wrong in either direction: Goo​gle may have both names reserved, or neither, and the forecast list may be recycling a name the account saw somewhere it no longer trusts. But between "a Flash Lite tier called Helium is next" and "a model called Helix is next," only the first has a post behind it that names a tier.

What a Gem​ini 4 Flash Lite would cost

Nothing is known about the price, so the honest exercise is to bound it with the models either side of the slot. All figures below are Goo​gle's published list prices as of 19 November 2026, not estimates.

• Gemini 3.5 Flash-Lite — $0.30 input / $2.50 output, $0.03 cached input, free tier available. The incumbent, and the number a successor has to beat or match.

• Gem​ini 4 Hydrogen — $1.50 / $7.50, $0.075 cached, free tier. The current cheap Gem​ini 4 tier, live since 12 November.

• Gemini 4 Carbon and Gemini 4 Argon — $2.00 / $10.00, $0.10 cached, intro rate through 30 November 2026, then $4.00 / $20.00 from 1 December.

• Gem​ini 4 Neon — $3.00 / $15.00, $0.15 cached, no free tier. The most expensive tier in the family today.

• The shape of the gap — if a Gem​ini 4 Flash Lite lands anywhere near the incumbent's $0.30/$2.50 it will be roughly five times cheaper on input than Hydrogen and, from 1 December, more than thirteen times cheaper than Carbon. That is the decision it would change, not the benchmark it would win.

Google AI Studio Gemini Developer API pricing page for Gemini 4 Carbon, showing the paid-tier rate card in USD per 1M tokens: input price $2.00 through November 30 2026 then $4.00 from December 1 2026, output price including thinking tokens $10.00 then $20.00, context caching price $0.10 then $0.20, the note that implicit caching is the only supported mode, and a free tier marked not available.

Two cautions on those numbers. Carbon and Argon doubling on 1 December is a scheduled hike on a live model, not a leak — anyone budgeting against the $2/$10 rate for December is already wrong. And a Flash Lite tier in the Gem​ini 4 line will almost certainly inherit the family's implicit-caching-only behaviour, since no Gem​ini 4 model supports explicit caching today; a $0.30-class model without an explicit cache is a different economics story than one with it.

What to do before it ships

Nothing. There is no model to call, no preview to queue for, no waitlist published, and no identifier to point a client at. Anything you build today against the string "gemini-helix" is building against a typo.

What you can do is put the tier you would actually replace on a routing layer now, so the swap is a config change rather than a migration. On OrcaRouter the models in this comparison that we route are google/gemini-3.5-flash-lite at the same $0.30/$2.50 Goo​gle charges, and google/gemini-4-carbon and google/gemini-4-argon at $2.00/$10.00. We pass through provider list price at 0% markup, which has one practical consequence for this specific story: when Carbon and Argon step to $4.00/$20.00 on 1 December, our rates move with Goo​gle's on the day, with no renegotiation and no separate contract. A price change is not a migration event.

The other use of a router in a leak cycle is the safe trial. When a new cheap tier does appear, the first week of it is unproven — capacity varies, rate limits move, and a model that benchmarks well on day one can be withdrawn or re-pointed by day ten. Sending a share of traffic to it with automatic failover onto the incumbent Flash Lite or onto Hydrogen means the experiment cannot take production down with it, and the question "is this good enough to switch" gets answered on real traffic instead of on a leaderboard.

We do not route Gem​ini 4 Hydrogen or Gem​ini 4 Neon, so if those are what you need, go to Goo​gle directly. Helix and Helium are not on OrcaRouter either, for the simpler reason that they do not exist as models anyone can call — the model catalogue is the list we can actually serve, and it will say so when that changes.

What would prove this wrong

Four things, in rough order of how fast they would resolve it.

A model ID appearing in the Gem​ini API — a gemini-4-helium or gemini-4-flash-lite string that returns something other than a 404. That is the only piece of evidence that ends the argument, and it is the evidence this leak does not have.

A fifth card on the Gem​ini API models page. Today there are four Gem​ini 4 entries and the page is the reference every SDK and client reads; a Flash Lite tier would have to appear there before anything could call it.

A pricing line. Goo​gle has announced every Gem​ini 4 tier with a rate card attached on day one, including the introductory and scheduled-increase dates for Carbon and Argon. A launch with no price would be the first of its kind in this family.

And on the other side: if 26 November passes with nothing, the "next week?" call has simply expired, which is what happens to most forecast lines and is not evidence of anything except that the account was guessing. Watch the 26 November forecast for whether the item is still listed, renamed, or quietly dropped — that, not the original post, is where the useful signal is.

The name is the story and the name is also the weakest part of it. Helium fits a naming pattern four releases deep, is attached to a tier that genuinely has not been refreshed, and is the only low-numbered element left unclaimed. Helix fits nothing except a keyboard. If a Gem​ini 4 Flash Lite ships in the next fortnight it will be called Helium, and the forecast that called it Helix will be remembered as the day the fastest account in this space mistyped its own scoop.