Is Gemini 4 Hydrogen free? Yes, on the AI Studio / API free tier and in the Gemini app. Free guide & student terms →
Gemini 4 Hydrogen comparisons: price, speed and cost per task
New Our original tests — playable same-prompt runs
Use cases Make a 3D game with AI · AI floor plan from video · AI 3D scene generator · Screenshot to HTML · AI promo video from code
New Same prompt, side by side — community time & cost
Comparison hub: every model at a glance
| Model | Official list price (in / out per 1M) | AA Index (best level) | Output tok/s (best) | AA cost per Index task | Comparison |
|---|---|---|---|---|---|
| Gemini 4 Hydrogen | $1.50 / $7.50 flat | 54 (xhigh) | 178 | $3.16 (xhigh) | Cost calculator |
| Gemini 3.8 Flash | $0.75 / $3.75 through 31 Dec 2026; $1.50 / $7.50 from 1 Jan 2027 | 31 (high) | 114 | $7.77 (high) | Hydrogen vs Gemini 3.8 Flash |
| Gemini 4 Argon | $2 / $10 through 30 Nov 2026; $4 / $20 from 1 Dec 2026 | 50 (high) | 66 | $4.97 (high) | Hydrogen vs Gemini 4 Argon |
| Gemini 4 Carbon | $2 / $10 through 30 Nov 2026; $4 / $20 from 1 Dec 2026 | 60 (high) | 66 | $5.48 (high) | Hydrogen vs Gemini 4 Carbon |
| Gemini 3.1 Pro Preview | $2 / $12 (≤200K prompt); $4 / $18 above | — | 113 | — | Hydrogen vs Gemini 3.1 Pro Preview |
| Claude Sonnet 5.5 | $2 / $10 | 55 (max) | 144 | $10.02 (max) | Hydrogen vs Claude Sonnet 5.5 |
| Claude Opus 5.5 | $4 / $20 | 59 (max) | 114 | $10.04 (max) | Hydrogen vs Claude Opus 5.5 |
| GPT-6.1 Sol | $2 / $10 (≤272K); $4 / $15 above | 55 (max) | 49 | $1.27 (max) | Hydrogen vs GPT-6.1 Sol |
Gemini 4 model comparisons beyond Hydrogen
- Gemini 4 Argon vs Carbon: benchmarks, pricing and speed
- Gemini 4 Carbon vs Claude Opus 5.5: coding, benchmarks and cost
Output tokens Artificial Analysis recorded while running its whole Intelligence Index (more tokens = more cost and time for the same work): Gemini 4 Hydrogen: xhigh 304M, high 110M, medium 54.4M, low 20.3M; Gemini 3.8 Flash: high 126M, medium 83.9M; Gemini 4 Argon: high 113M, medium 81.7M; Gemini 4 Carbon: high 205M, medium 124M, low 83.9M; Claude Sonnet 5.5: max 335M, xhigh 111M, high 54.7M, medium 29.4M, low 21.3M; Claude Opus 5.5: max 218M, xhigh 101M, high 56.1M, medium 42.5M, low 21.1M; GPT-6.1 Sol: max 63.7M, xhigh 33.3M, high 24.7M, medium 15.1M, low 8.18M.
Independent measurements: Artificial Analysis: Gemini 4 Hydrogen · Artificial Analysis: Gemini 3.8 Flash · Artificial Analysis: Gemini 4 Argon · Artificial Analysis: Gemini 4 Carbon · Artificial Analysis: Gemini 3.1 Pro Preview · Artificial Analysis: Claude Sonnet 5.5 · Artificial Analysis: Claude Opus 5.5 · Artificial Analysis: GPT-6.1 Sol (retrieved 14 November 2026).
Calculate your cost per task → · Errors & limits FAQ
Quick answers vs 3.8 Flash Benchmarks vs Argon / Carbon vs Sonnet 5.5 vs Opus 5.5 vs GPT-6.1 Sol vs 3.1 Pro Pricing Cost calculator FAQ Errors & limits
Two kinds of figures appear below, and they should not be mixed. First, each company’s own list prices and context-window specifications, retrieved on 13 November 2026. Second, the benchmark table Google published on the Gemini 4 Hydrogen model card in November 2026, which reports Google’s measurements of Hydrogen, Claude Sonnet 5.5, and GPT-6.1 Terra. This site has not rerun those evaluations.
Official list prices and context windows
Prices are standard / short-context API list prices unless a vendor uses a long-context surcharge. Hydrogen is priced flat across its full 1M window on the pages we read. GPT-6.1 Terra and grok-4.7 are not: they charge more above a token threshold. Claude Sonnet 5.5 is listed at a flat 1M-window rate.
Source: Google AI Studio / Gemini API docs, Retrieved 13 November 2026 · Source: OpenAI, Retrieved 13 November 2026 · Source: Anthropic, Retrieved 13 November 2026 · Source: xAI, Retrieved 13 November 2026
| Model | Vendor | Input / 1M | Output / 1M | Cache read / 1M | Context / max output |
|---|---|---|---|---|---|
| Gemini 4 Hydrogen gemini-4-hydrogen | $1.50 flat | $7.50 flat | $0.075 implicit | 1,048,576 unified; 65,536 sync hop, up to 1M via long decode | |
| Gemini 4 Carbon / Argon | $2.00 to 30 Nov 2026; $4.00 from 1 Dec | $10.00 / $20.00 | $0.10 / $0.20 | 1,048,576 unified; no xhigh; no free tier | |
| Gemini 3.8 Flash | $0.75 to 31 Dec 2026; $1.50 from 1 Jan 2027 | $3.75 / $7.50 | $0.075 / $0.15 + storage | 1,048,576 input; 65,536 output | |
| GPT-6.1 Terra gpt-6.1-terra | OpenAI | $1.00; $2.00 if prompt > 272K | $5.00; $7.50 if prompt > 272K | $0.05; $0.10 long | 1,050,000 context; 922,000 max input; 128,000 max output |
| Claude Sonnet 5.5 claude-sonnet-5-5 | Anthropic | $2.00 flat | $10.00 flat | $0.10 | 1M context; 128K max output (300K on Batches API, beta) |
| grok-4.7 | xAI | $2.00; $4.00 if prompt ≥ 200K | $6.00; $12.00 if prompt ≥ 200K | $0.50; $1.00 long | 500k context (vendor pricing table) |
Google’s published Hydrogen benchmarks
Copied from the Gemini 4 Hydrogen model card, results as of November 2026. Methodology: DeepMind evaluation methodology. A dash means the card left that cell blank. “Leads its class” and similar phrases from the announcement blog are Google’s wording, not ours.
Source: Google DeepMind, 12 November 2026
| Category | Benchmark | Notes | Gemini 4 Hydrogen | Claude Sonnet 5.5 | GPT-6.1 Terra |
|---|---|---|---|---|---|
| Agentic coding | SWE-sweep | 13.5% | 3.1% | 5.8% | |
| Terminal-Bench 4.1 | Cheating-adjusted | 57.0% | 63.9% | — | |
| Knowledge work | Agents’ Last Exam v1.1 | Score | 71.6% | 60.4% | 69.8% |
| AutomationBench-AA | 76.1% | 71.8% | 60.2% | ||
| Vending-Bench 2 | Bank balance ($) | $14,549 | $11,165 | $9,107 | |
| Computer use | OSWorld-2.1 | Offline subset | 77.6% | 83.9% | 74.2% |
| Multimodal | MKA | Climb WSR accuracy | 53.4% | 52.0% | 43.8% |
| Chartography | 66.6% | 61.6% | 63.2% | ||
| Science and math | Terminal-Bench-Science 0.1 | 51.4% | 59.9% | 58.3% | |
| Reasoning | HLE-Diamond | 58.8% | 53.3% | 47.3% |
A separate official number from Anthropic
Anthropic’s 28 September 2026 Sonnet 5.5 announcement reports 70.6% on Terminal-Bench 4.0 (and 63.9% is not the number used on that page). Because the edition differs from Google’s 4.1 table, we do not place 70.6% in the table above. Source: Anthropic, 28 September 2026.
Other official claims (not a single scoreboard)
- Artificial Analysis Intelligence Index: Hydrogen 51 at default high thinking, 54 at xhigh, versus 50 for Argon. (Google announcement blog, 12 November 2026.)
- Terminal-Bench 4.1 versus Gemini 3.8 Flash: Hydrogen 57% vs 18.2%, and “finishing tasks 2.5 times faster at its high default setting.” (Same blog.) We have not independently timed the models.
- The same blog says Hydrogen “cuts fabricated claims by more than 5x” versus 3.8 Flash, reduces overreach in agentic tasks by up to 81%, and names several in-class leads (HLE-Diamond 58.8%, Vals Tax Agent Bench 52.3% as #1, Vals Legal Research Bench 55.8% tied for #1, MKA 53.4%, Agents’ Last Exam v1.1 71.6%). Those extra named scores are from the blog; some also appear in the model-card table.
- The Carbon model card (results as of October 2026) reports a different, earlier table that includes Carbon, Argon, GPT-6 Astra, Claude Opus 5.5, and Claude Fable 5.1, and uses Terminal-Bench 4.0 rather than 4.1. It is not directly the same scoreboard as the Hydrogen card. Carbon model card.
- Google’s Hydrogen table compares against GPT-6.1 Terra, not GPT-6.1 Sol. Autocomplete interest in a Sol comparison is real; a Hydrogen-versus-Sol official table from Google was not on the pages we read, so we do not invent one.
- xAI’s grok-4.7 appears in the price table because official list prices are published. Google’s Hydrogen model card does not report grok-4.7 scores, so none are shown here.
Source: Google Blog, 12 November 2026