Gemini 4 Hydrogen Costs 2x Gemini 3.8 Flash Per Token. Per Task, It Costs $1.87 to Flash’s $7.77 and Scores 51 to 31.
Google priced Gemini 4 Hydrogen, released November 12, at $1.50 per million input tokens and $7.50 per million output, twice Gemini 3.8 Flash. On the Artificial Analysis Intelligence Index it costs $1.87 per task to Flash’s $7.77 and scores 51 to 31. The catch: GPT-6.1 Sol scores 50 on the same index for $0.59.
The Flash that dropped the name
Hydrogen replaces Gemini 3.8 Flash as Google’s fast, cheap tier, and it is the first model in that line without “Flash” in the name. Google pitches it for coding and agent work that needs speed, and leaves long, hard reasoning to its frontier models, Gemini 4 Carbon and Gemini 4 Argon.
The model ID is gemini-4-hydrogen. It reads up to 1M tokens, writes up to 1M, and has 4 thinking levels: low, medium, high (the default) and a new xhigh. It’s rolling out in the Gemini API, AI Studio, the Gemini app, Google Search and Gemini for business. In Google Antigravity it goes to paid Google AI Pro and Ultra subscribers; we covered what Antigravity’s free tier kept in Google Pulled Claude From Antigravity’s Free Tier.
The price sheet: 2x today, equal on January 1
| Per 1M tokens, paid tier | Gemini 4 Hydrogen | Gemini 3.8 Flash to Dec 31 | Gemini 3.8 Flash from Jan 1, 2027 |
|---|---|---|---|
| Standard input / output | $1.50 / $7.50 | $0.75 / $3.75 | $1.50 / $7.50 |
| Cached input | $0.075 | $0.075 | $0.15 |
| Batch and Flex input / output | $0.75 / $3.75 | $0.375 / $1.875 | $0.75 / $3.75 |
| Priority input / output | $2.70 / $13.50 | $1.35 / $6.75 | $2.70 / $13.50 |
| Free tier | Yes (Standard, Priority) | Yes | Yes |
Google AI Studio pricing page. Output prices include thinking tokens. Hydrogen supports implicit caching only; Flash also bills $0.50 per million tokens per hour for cache storage, rising to $1.00 on January 1.
Two lines stand out. Hydrogen’s cached input costs the same $0.075 as Flash’s today, so an agent that rereads the same long context pays no more for those tokens. And on January 1, Flash’s standard price doubles to exactly Hydrogen’s. From that day, Flash costs the same per token and scores 20 points lower on the index.
Flash isn’t going away. Google’s deprecations page lists no shutdown date for gemini-3.8-flash, and on December 1 it becomes the automatic target for requests to gemini-2.5-pro.
Why the pricier model is cheaper per task
Artificial Analysis runs every model through the same Intelligence Index and bills the run at list prices, cache hits included. Here is what it recorded at each model’s high setting:
| Intelligence Index run, high | Gemini 4 Hydrogen | Gemini 3.8 Flash |
|---|---|---|
| Index score | 51 | 31 |
| Cost per task | $1.87 | $7.77 |
| Input tokens | 5.34B | 10.2B |
| Output tokens | 110M | 126M |
| Cache discount | 95% | 90% |
| Output speed | 174 tokens/s | 113 tokens/s |
| Time per task | 8.1 min | 14.6 min |
Artificial Analysis model pages and leaderboard, Intelligence Index v5.0.1.
Hydrogen read about half the input tokens Flash did and wrote 13% fewer output tokens, so a 2x token price still came out 76% cheaper per task. It also finished each task in 55% of the time. Google’s own numbers point the same way: it says Hydrogen finishes tasks 2.5x faster than 3.8 Flash at the default setting, and that partners in early testing saw runs 2x to 2.8x faster with 12% to 23% fewer tokens.
For agent loops, the token price is the wrong number to compare. Flash’s cheaper tokens bought a run that cost 4.2x more and scored 20 points lower.
On coding, Google reports 57% on Terminal-Bench 4.1 for Hydrogen to 18.2% for 3.8 Flash. On the Intelligence Index, Hydrogen at high (51) edges out Gemini 4 Argon at high (50), which costs $4.97 a task.
The dial: 4 settings, $0.29 to $3.16
Artificial Analysis ran Hydrogen at all 4 thinking levels. Next to it, the models a builder would weigh at each price point:
| Model and setting | Intelligence Index | Cost per task |
|---|---|---|
| Gemini 4 Hydrogen (xhigh) | 54 | $3.16 |
| Gemini 4 Hydrogen (high) | 51 | $1.87 |
| Gemini 4 Hydrogen (medium) | 41 | $0.80 |
| Gemini 4 Hydrogen (low) | 27 | $0.29 |
| Gemini 3.8 Flash (high) | 31 | $7.77 |
| Gemini 3.8 Flash (medium) | 30 | $6.26 |
| GPT-6.1 Sol (max) | 55 | $1.27 |
| GPT-6.1 Sol (high) | 50 | $0.59 |
| GPT-6.1 Sol (medium) | 46 | $0.36 |
| Claude Opus 5.5 (high) | 52 | $3.46 |
| Claude Sonnet 5.5 (xhigh) | 50 | $4.09 |
| Gemini 4 Carbon (high) | 60 | $5.48 |
| Gemini 4 Argon (high) | 50 | $4.97 |
| Claude Haiku 5.5 (max) | 36 | $0.41 |
| Muse Ember (max) | 36 | $0.14 |
| GPT-6 Luna (max) | 28 | $0.15 |
Artificial Analysis leaderboard, Intelligence Index v5.0.1.
Against Google’s own lineup, Hydrogen wins. It beats Argon on the index for 38% of the cost per task, and 3.8 Flash at every setting.
Against GPT-6.1 Sol, it doesn’t. Sol at high scores 50 for $0.59, a third of Hydrogen’s $1.87 for 51. Sol at max scores 55 for $1.27, more points than Hydrogen at xhigh for 40% of its $3.16. Sol at medium (46 for $0.36) beats Hydrogen at medium (41 for $0.80).
At the low end, small models win. Hydrogen at low scores 27 for $0.29. Muse Ember scores 36 for $0.14 and Haiku 5.5 scores 36 for $0.41.
When per-token price still decides
Per-task numbers matter for agent loops, where the model decides how many tokens to burn. For short calls with no or little reasoning, such as tagging, extraction or captions, your token count is roughly fixed and the price sheet is the bill. Our arithmetic for 1,000 calls of 4,000 input and 500 output tokens, no cache:
| 1,000 calls, 4,000 in / 500 out | Standard | Batch or Flex |
|---|---|---|
| Gemini 3.8 Flash, to Dec 31 | $4.875 | $2.4375 |
| Gemini 4 Hydrogen | $9.75 | $4.875 |
| Gemini 3.8 Flash, from Jan 1 | $9.75 | $4.875 |
Our math from Google’s list prices, assuming both models use the same tokens. Thinking tokens bill as output and would raise every row.
So for about 7 more weeks, Flash is the cheaper choice for fixed-size, high-volume calls, and Hydrogen on Batch or Flex costs what Flash on Standard does now. After January 1 the gap closes, and the only reason left to stay on Flash is that your prompts are tuned for it.
What this means if you build with AI
Agent loops on Gemini: switch to Hydrogen. If you run coding or browsing agents on 3.8 Flash, the index says you pay 4.2x more per task for 20 fewer points. Hydrogen at high is the new Gemini default for this work, and medium ($0.80, 41) is the setting to try when cost matters more than the hard cases.
Cheapest good agent model overall: still GPT-6.1 Sol. At high it scores 50 for $0.59 a task. If you pick models by cost per task and don’t need Google’s stack (Search grounding, AI Studio, Vertex), Hydrogen doesn’t take that slot.
Bulk short calls: Flash until December 31, then re-run the math. Move anything that can wait to Batch or Flex for half price on either model.
Budget the January 1 change now. A pipeline that stays on 3.8 Flash will cost 2x per token on New Year’s Day with no code change. Put the model name in config, not code, so the swap is one line.
On MegaBrain, the models from this post that you can call today through one OpenAI-compatible API and one balance include GPT-6.1 Sol, Gemini 3.8 Flash, Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5; Gemini 4 Hydrogen isn’t in the catalog yet. If you run a coding agent such as OpenCode on that key, the OpenCode setup guide shows the config, and GPT-6.1 Sol is the per-task value pick from the table above.
Sources
Pages opened November 14, 2026. Every number above comes from these pages, or from our arithmetic on them where we say so.
- Google — Introducing Gemini 4 Hydrogen for Agentic Workflows (November 12, 2026: price, speed, context, thinking levels, benchmarks, model ID)
- Google AI Studio — Gemini API pricing (Gemini 4 Hydrogen and Gemini 3.8 Flash: Standard, Batch, Flex, Priority, and the January 1, 2027 change)
- Google AI Studio — Gemini API deprecations (gemini-3.8-flash: no shutdown date; gemini-2.5-pro routing)
- Google Antigravity — Gemini 4 Hydrogen and Orchestrator (plans, Terminal-Bench 4.1, Orchestrator quota and speed)
- Artificial Analysis — LLM leaderboard (Intelligence Index v5.0.1 and cost per task by reasoning setting)
- Artificial Analysis — Gemini 4 Hydrogen (high) model page (tokens, speed, time per task)
- Artificial Analysis — Gemini 3.8 Flash model page (tokens, speed, time per task)
Full disclosure: MegaBrain sells pay-as-you-go API access to image, video and text models, including Gemini 3.8 Flash, GPT-6.1 Sol, Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 5.5. MegaBrain is not affiliated with Google, OpenAI, Anthropic, Meta or Artificial Analysis.
MegaBrain for Mac · an AI coworker
Give MegaBrain the task, not the steps.
It works on your files, writes documents, spreadsheets and decks, connects to your team’s tools and asks before anything risky. Free to start.
Newsletter
Stay in the loop
Get our latest AI research and guides. No spam, unsubscribe anytime.