Gemini 4 API: Latest Release Status, Pricing and What Developers Can Use Today
Track Gemini 4 Pro checkpoints argon and barium-b, LMArena tests, the reported 256K output and 2M context target, API availability, release date and pricing.
Published 2026-09-17 · Updated 2026-09-26
For the announced Argon model and its restricted-access status, see the Gemini 4 Argon deep dive.
Gemini 4 is not publicly available as an API as of September 26, 2026. The most specific current signal is a pair of internal checkpoints observed in LMArena, not a public Gemini 4 model ID, model card, API price table, release date, or APIMaster route. Google's public Gemini product line still consists of released products such as Gemini 3.1 Pro and Gemini 3.8 Flash.
APIMaster price check, 2026-09-24: The table puts the reference price, current APIMaster price, active routes, and model card side by side. Prices can change; use the market card as the source of truth。
| Model ID | Reference price (per 1M input / output) | Current APIMaster price | Active routes | Discount | Model card |
|---|---|---|---|---|---|
gemini-3.8-flash |
$1.50 / $7.50 |
$0.2999 / $1.4993 |
7 | 80% off | View discount |
gemini-3.7-flash |
$1.50 / $7.50 |
$0.2999 / $1.4993 |
7 | 80% off | View discount |
Google has publicly confirmed that Gemini 4 is in training. On Alphabet's July 2026 earnings call, Sundar Pichai said: “We are now training Gemini 4, and we’re being very ambitious with it.” He also said he was excited by the internal progress and expected people to be pleased when Google puts it outside. Google is aiming to compete with the frontier available when Gemini 4 launches. That confirms active development, not an API announcement. Until Google publishes a named model in its API documentation, developers should not send production traffic to guessed identifiers such as gemini-4 or assume that Gemini 3.8 capabilities, context limits, or prices carry over.
What can developers use today? Gemini 4 has no APIMaster route, so the practical choice is the released Flash tier: Gemini 3.8 Flash is live on APIMaster from about $0.20 input / $1.00 output per 1M tokens, roughly 87% below Google's $1.50 / $7.50 standard list price. Check the live Gemini 3.8 Flash card for the routes on sale today, or create an account and test it against your own workload. This tracker will be updated on the same URL when Google publishes official release, API, pricing, and model-card details.
First published September 17, 2026; updated and checked September 26, 2026. APIMaster route prices were re-checked September 24, 2026.
GPT Trial(GPT-6 Astra, GPT-6 Sol, GPT-5.6, and GPT-5.5 available)
Sign up and claim a $20 GPT trial
Gemini 4 status at a glance
| Question | Status as of September 26, 2026 |
|---|---|
| Is Gemini 4 publicly released? | No public release found |
| Is there a confirmed Gemini 4 API model ID? | No |
| Is Gemini 4 listed in Google's public API model documentation? | No public listing found |
| Is there an official Gemini 4 price? | No |
| Is there an official launch date? | No |
| Is Gemini 4 in training? | Yes; Sundar Pichai confirmed it on Alphabet's July 2026 earnings call |
| What is publicly available now? | Gemini 3.8-family products and routes, subject to the current vendor and provider catalog |
Gemini 4 Pro checkpoints: argon, barium-b and LMArena
The most concrete Gemini 4 Pro signal currently visible in the community is not a public API release. It is a sequence of internal checkpoints tested in LMArena. These checkpoints have commonly appeared under labels such as gemini-3.8-flash or gemini-3.7-flash, so the visible Arena label should not be treated as proof of the underlying model identity or as a stable API model ID.
The earlier checkpoint known as argon appeared around September 14, 2026. A newer checkpoint called barium-b appeared around September 25. In the community's current reading, the progression from argon to barium-b is a meaningful testing signal, while neither name represents a public Google product or an APIMaster route.
| Reported checkpoint signal | Current community observation | What developers should conclude |
|---|---|---|
| Output limit | About 256K tokens | A checkpoint observation, not a published API limit |
| Context target | 2M tokens is a reported target | Not confirmed; do not budget for it yet |
| Reasoning mode | High thinking effort is visible in tests | Long runs can take several minutes to more than ten minutes |
| Strong tasks | Front-end and web design, SVG, 3D/voxel scenes, interactive UI | Promising demos, not a standardized benchmark or product guarantee |
Community demos include full-site generation, an Xbox controller SVG, a helicopter 3D model, and voxel towers. They suggest a notable improvement in visual implementation and interaction design over the template-like output associated with earlier Gemini generations. Google appears to be moving from Gemini 3.1 Pro toward Gemini 4 without a clearly public Gemini 3.5 Pro product, but that product-line reading is still community interpretation. The formal API products available today remain Gemini 3.1 Pro, Gemini 3.8 Flash, and other entries in Google's live catalog.
What is confirmed, and what is not?
The current public baseline: Gemini 3.8
Google's current public Gemini documentation and product pages are the source of truth for usable API models. They provide the released Gemini 3.8 family, its supported features, and the API-specific details that developers need before integration. A newer internal training run does not replace that documentation.
For an application owner, the safe production rule is simple: use only a model name that appears in the official API catalog or a live provider catalog, then verify it with a real request. A blog post, a benchmark screenshot, a product label, or a route nickname does not establish a stable API contract.
Gemini 4 training is officially confirmed
On Alphabet's July 2026 earnings call, Sundar Pichai said: “We are now training Gemini 4, and we’re being very ambitious with it.” He added: “I am very excited by the progress I’m seeing internally… people will be pleased when we are putting it outside.” The stated target is the frontier available when Gemini 4 is released, rather than the frontier available when the remarks were made.
This is the strongest public signal in the tracker: the CEO has confirmed active Gemini 4 training and expressed confidence in internal progress. It is still not a public product specification. The call did not publish an API model ID, a product date, a model card, benchmarks, context limits, or pricing.
Product-line signals from Logan Kilpatrick
Logan Kilpatrick, who works closest to the Gemini API and AI Studio product line, has reinforced the training signal in public posts:
- July 21, 2026: “We have started our most ambitious pre-training run yet, for Gemini 4.”
- August 14, 2026: he repeated that Gemini 4 is Google's most ambitious pre-training run to date.
- September 10, 2026: he added that Gemini 4 models will continue to push the frontier on cyber.
These are useful product-direction statements. They support the conclusion that Gemini 4 is an active, ambitious training program and that cyber capability is an explicit direction. They do not confirm a public cyber product, API capability, benchmark score, safety policy, or availability date.
That still leaves the facts developers actually need unknown: the final name, release date, model variants, supported modalities, context window, tool support, reasoning controls, safety policy, API endpoint, model ID, regional availability, and price. Training-stage checkpoints can change materially before public release.
What is not confirmed
- A public Gemini 4 API endpoint or model ID.
- A launch date, including an October window mentioned in secondary discussion.
- Whether Gemini 4 will ship as Pro, Flash, Live, or another family of variants.
- Benchmarks, context limits, output limits, tool support, or computer-use capabilities.
- Token prices, cached-token prices, batch prices, or APIMaster availability.
- Claims that Gemini 3.5 Pro has been retired or that all of its compute has moved to Gemini 4.
Gemini 4 release timeline
| Date | Signal | What it establishes |
|---|---|---|
| July 2026 | Sundar Pichai said on Alphabet's earnings call: “We are now training Gemini 4, and we’re being very ambitious with it” | Official confirmation that Gemini 4 is in training; no API specification or date |
| July 21, 2026 | Logan Kilpatrick said Google had started its most ambitious pre-training run yet for Gemini 4 | Product-line confirmation of the training program |
| August 14, 2026 | Kilpatrick repeated the ambitious-pre-training characterization | The training program remained an active public direction |
| September 2, 2026 | Google announced Gemini 3.8 Flash | The released 3.8 family is the public developer baseline |
| September 10, 2026 | Kilpatrick said Gemini 4 models would continue to push the frontier on cyber | Cyber is a stated direction; no public Gemini 4 cyber model or API has been announced |
| September 14, 2026 | The argon checkpoint was reported in LMArena testing |
A concrete internal-test signal; not a public API model |
| September 17, 2026 | Google API documentation and public product pages still expose the Gemini 3.8 line rather than Gemini 4 | No public Gemini 4 API contract, pricing, or model card found |
| September 25, 2026 | The newer barium-b checkpoint was reported, with more front-end and 3D demos circulating |
A newer checkpoint signal; the Arena alias still does not prove a public Gemini 4 identity |
| September 26, 2026 | Community reports describe about 256K output, a possible 2M context target, and High thinking effort | Useful test observations, not official specifications |
| Future | Official Google model page, API catalog entry, price table, or release note | The evidence required to mark Gemini 4 as available |
The key distinction is between a model in development and a model developers can integrate. The second requires a public model identifier and documentation. This page will update the release row only when that evidence is available.
Gemini 4 API availability
There is no confirmed public Gemini 4 API model ID. Do not configure gemini-4, gemini-4-pro, gemini-4-flash, or any similar guessed identifier in production. A provider may expose an experimental label before Google's public documentation is complete, but the label alone does not prove the model identity, feature set, availability guarantees, or billing behavior.
When a Gemini 4 route becomes visible, verify these items before moving traffic:
- The exact model ID returned by a successful API response.
- The API endpoint, streaming behavior, supported inputs, structured output, function calling, search tools, and any computer-use capability.
- Context limits, maximum output, rate limits, regional availability, safety behavior, and deprecation policy.
- Input, cached-input, output, long-context, batch, and tool-related charges.
- Route identity and quality on a saved evaluation set, including retries, latency, and total accepted-task cost.
APIMaster will list a route only when it is actually available, so there is nothing to buy under a Gemini 4 label today. What you can buy now is the current generation: Gemini 3.8 Flash from $0.20 / $1.00 per 1M tokens. Use the AI API model tester to validate a route before committing production usage.
Gemini 4 Pro: what the Pro tier would mean
Google has not announced a "Gemini 4 Pro". The name follows the way the Gemini line has been tiered so far — Pro for the higher-capability, higher-cost class, Flash for the faster, lower-cost class — and it is a reasonable expectation rather than a confirmed product. Treat every Gemini 4 Pro release date, model ID and price you see today as unverified until Google publishes it.
| Question | Answer today |
|---|---|
| Has Google announced Gemini 4 Pro? | No official announcement found |
| Is there a Gemini 4 Pro model ID? | No |
| Is there a Gemini 4 Pro price? | No |
| Is there a Gemini 4 Pro release date? | No |
| Can I call it on APIMaster? | No route under that name |
If Google keeps the naming used through the 3.x generation, a Gemini 4 launch would most likely arrive as a Pro-class model plus a Flash-class model. That is a pattern, not a commitment: confirm the model card and API documentation instead of assuming a Pro variant exists on day one, or that it keeps today's context limits and price relationship.
What you can compare right now is the released Flash tier. gemini-3.8-flash starts at $0.20 input / $1.00 output per 1M tokens on APIMaster, about 87% below Google's $1.50 / $7.50 standard list price (checked September 20, 2026) — see the Gemini 3.8 Flash card for the routes on sale. If you are specifically waiting on a Pro-class Gemini 4, this page will record its model ID, price table and APIMaster route as soon as they are published.
Gemini 4 pricing
Google has not published Gemini 4 pricing. There is no reliable input-token, output-token, cache, batch, or long-context rate to quote. Existing Gemini 3.8 prices are not a Gemini 4 estimate and should not be carried forward into a budget or purchase decision.
| Pricing question | Gemini 4 answer today |
|---|---|
| Official input-token price | Not published |
| Official output-token price | Not published |
| Cached-input price | Not published |
| Batch or long-context pricing | Not published |
| APIMaster route price | No confirmed Gemini 4 route |
While Gemini 4 has no route, the generation that is on sale does: gemini-3.8-flash at $0.20 in / $1.00 out and gemini-3.7-flash at the same rate, against Google's $1.50 / $7.50 list price — see the Gemini 3.8 Flash card and the Gemini 3.7 Flash card for current routes.
When Gemini 4 pricing is announced, compare the official rate with the live APIMaster marketplace rate separately. Provider routes can differ by availability, channel, capacity, and billing terms. The useful comparison is cost per accepted task, not only cost per million tokens.
Gemini 4 vs Gemini 3.8: what can be compared now?
| Area | Gemini 3.8 | Gemini 4 |
|---|---|---|
| Public release | Released | Not publicly released |
| API model ID | Published for released variants | No confirmed ID |
| API documentation | Available for released variants | No public documentation found |
| Price | Published for released variants | Not published |
| Model card | Available for released variants where provided by Google | Not published |
| Capabilities | Must be checked per released variant | Not confirmed; do not infer from 3.8 |
| APIMaster availability | Check the live marketplace | No confirmed route |
Gemini 3.8 is a useful reference for current workloads, but it is not a substitute for a Gemini 4 specification. Do not assume a Gemini 4 release will keep the same Flash or Pro naming, model behavior, API surface, or price relationship.
What developers should prepare before launch
The productive work before a new model ships is to make the evaluation path ready, not to guess the final model ID. Keep prompts, tool definitions, input files, grading criteria, and logging stable so Gemini 4 can be compared with the current Gemini route on its first day of public availability.
| Workload | Measure after Gemini 4 is released |
|---|---|
| Coding agent | Tests passed, regressions, repair turns, completion time, and total cost |
| Tool-using workflow | Successful tool sequence, recovery from errors, tool-call volume, and p95 latency |
| Multimodal analysis | Extraction accuracy, citation quality, unsupported claims, and cost |
| Long-context task | Recall, missed constraints, context handling, output length, and bill |
| High-volume requests | Throughput, error rate, latency, and cost per accepted output |
This approach will establish whether Gemini 4 is a real improvement for a specific workload rather than relying on an early claim or a single demo.
FAQ
What is barium-b?
barium-b is a reported internal checkpoint tested in LMArena around September 25, 2026. It is not a public model ID, an official Gemini 4 Pro release, or an APIMaster route. The Arena may display it under a Gemini 3.8 Flash or Gemini 3.7 Flash label, which is a test alias rather than a product contract.
Is argon Gemini 4 Pro?
argon is an earlier checkpoint reported around September 14. Its behavior and lineage are consistent with the community's Gemini 4 Pro theory, but the name alone does not establish an official model identity. Treat it as an internal test checkpoint until Google publishes a model card and API documentation.
Will Gemini 4 support 256K output or 2M context?
Community testing reports about 256K-token output and a 2M-token context target. The output figure is an observed checkpoint ceiling, while 2M remains a target report. Neither should be treated as a guaranteed public API limit.
Is Gemini 4 released?
No public Gemini 4 release or API availability has been confirmed as of September 26, 2026. The argon and barium-b checkpoints are the strongest current community signals, but Google's released Gemini products remain the public baseline.
Is Gemini 4 already in training?
Yes. Sundar Pichai publicly confirmed on Alphabet's July 2026 earnings call that Google is training Gemini 4. Logan Kilpatrick also described it as Google's most ambitious pre-training run. Neither statement confirms a final product specification, release date, public model ID, API access, or price.
When will Gemini 4 be released?
Google has not announced a release date. An October window, often described as mid-October in community discussion, is not an official launch commitment.
What will the Gemini 4 API model ID be?
No model ID has been published. Do not treat names such as gemini-4, gemini-4-pro, or gemini-4-flash as usable API identifiers until they appear in official documentation and complete a real test request.
Is there a Gemini 4 Pro?
Google has not announced a model by that name. "Gemini 4 Pro" is an expected tier label based on how Google named previous generations, so there is no confirmed Gemini 4 Pro model ID, price or release date to track yet. The released tier you can call today is gemini-3.8-flash, from $0.20 / $1.00 per 1M tokens on APIMaster against a $1.50 / $7.50 reference list price (checked September 20, 2026) — see the Gemini 3.8 Flash card.
How much will Gemini 4 cost?
Gemini 4 pricing has not been published. Do not use Gemini 3.8 prices as a Gemini 4 forecast.
Can I use Gemini 4 on APIMaster now?
No confirmed Gemini 4 route is available through APIMaster at the time of this update. Check the live marketplace for released Gemini routes.
What should I use instead of Gemini 4 today?
Use a currently released Gemini model that appears in the official or live provider catalog. For the latest published Gemini reference point, see the Gemini 3.8 Flash guide.
Sources and verification watchlist
- Alphabet investor relations - July 2026 earnings-call remarks from Sundar Pichai; the cited transcript was also reported by INKL.
- Logan Kilpatrick - Gemini API / AI Studio product-line posts on July 21, August 14, and September 10, 2026.
- Google AI for Developers - Gemini models
- Google AI for Developers - Gemini API release notes
- Google DeepMind - Gemini models
- Google - Gemini 3.8 Flash announcement
- LMArena community observations reported by @Lentils80 and others around September 14, 2026 (
argon), and by @lyraxana and @thtbee_ around September 25, 2026 (barium-band related demos). These are checkpoint observations, not official Google specifications.
Gemini 4 training is confirmed by Google leadership, and argon/barium-b add a concrete internal-testing signal. Add a direct model card, API documentation, release note, and price table here before upgrading checkpoint observations to confirmed public product capabilities.