Gemini 4 Argon
Gemini 4 Argon is a proprietary model from Google with a 1.05M-token context window. 2 providers list it, all at $4.00 per million input tokens and $20.00 per million output tokens, checked 25 Oct 2026.
Providers serving Gemini 4 Argon
Both charge $4.00 in · $20.00 out. Cheapest first, by the blended rate of three input tokens to one output token. Prices in US dollars per million tokens, checked 25 Oct 2026; the source link opens where the price is published. Both list tool calling and JSON mode; neither lists an OpenAI-compatible API.
| Provider | Input | Output | Context | Model ID | Source |
|---|---|---|---|---|---|
| Google AI Studio1Maker's API | $4.00 | $20.00 | 1.05M | gemini-4-argon | |
| Google Vertex2 | $4.00 | $20.00 | 1.05M | gemini-4-argon |
- Standard price; $2/$10 until 2026-11-30.
- Standard price; a lower promotional price applies for a limited time.
How each provider handles your prompts
For each provider's paid API, from its own pages, checked 26 Oct 2026. Where a promise lives matters: a signed agreement binds the provider, while a docs page can change without notice. Not legal advice.
| Provider | Trains on your prompts | Keeps prompts | Zero retention | Data processing agreement |
|---|---|---|---|---|
Google AI StudioFree tier differs: Google uses free-tier prompts and responses to improve its products, with human review (paid-tier rules apply to free use in the EEA, Switzerland and UK). Grounding with Google Search or Maps stores prompts and output for 30 days, and this cannot be switched off.MoreLess | No | 55 days, for abuse monitoring | Not offered | Public |
Google VertexClaude Fable/Mythos keep all traffic 30 days and require sharing it with Anthropic; Google Search grounding keeps queries 3 days; the 24-hour cache can be switched off; the Interactions API stores prompts unless store=false.MoreLess | No | Not stored beyond the request, except a 24-hour cache (on by default) and abuse-flagged prompts (up to 90 days) | Enterprise plans only | Public |
Gemini 4 Argon benchmark scores
Compare every model in Bench rankingsPublished scores only, each linked to its source. Ticks are every other model's best published score.
GPQA Diamond
PhD-level scientific reasoning, pass@1, no tools
high effort 96.4%, source: Epoch AI, high effortRanks 1st of 221 models with a score
APEX-Agents
Agentic office work, APEX-Agents 1.1, Pass@1 in the maintainers' harness
high effort 82.4%, source: APEX-Agents leaderboard (Mercor), high effortRanks 1st of 48 models with a score
Humanity's Last Exam (text-only)
Reasoning & knowledge, no tools
high effort 57.1%, source: Artificial Analysis, high effortRanks 3rd of 221 models with a score
CritPt v2
Physics reasoning
72.9%, source: CritPt v2 leaderboardRanks 6th of 25 models with a score
SciCode (with background)
Scientific coding
high effort 61.8%, source: Artificial Analysis, high effortRanks 3rd of 118 models with a score
GDPval-AA v2.1 (Elo)
Real-world work tasks; a rating against other models, not a percentage
high effort 1624, source: Artificial Analysis, high effortRanks 17th of 144 models with a score
LCR
Long-context reasoning, AA-LCR v1.1
high effort 79.7%, source: Artificial Analysis, high effortJoint 43rd of 211 models with a score
Omni Accuracy
Omniscience knowledge accuracy
high effort 49.9%, source: Artificial Analysis, high effortRanks 20th of 216 models with a score
Non-Hallucination
1 − hallucination rate
high effort 84.9%, source: Artificial Analysis, high effortRanks 3rd of 216 models with a score
Each score is the model's best published result across its settings, named beside it. A setting above the default can use more tokens per task, which a per-token price does not show.
Show as table
| Benchmark | Score | Setting |
|---|---|---|
GPQA Diamond PhD-level scientific reasoning, pass@1, no tools | 96.4%, source: Epoch AI, high effort | high effort |
APEX-Agents Agentic office work, APEX-Agents 1.1, Pass@1 in the maintainers' harness | 82.4%, source: APEX-Agents leaderboard (Mercor), high effort | high effort |
Humanity's Last Exam (text-only) Reasoning & knowledge, no tools | 57.1%, source: Artificial Analysis, high effort | high effort |
CritPt v2 Physics reasoning | 72.9%, source: CritPt v2 leaderboard | Default |
SciCode (with background) Scientific coding | 61.8%, source: Artificial Analysis, high effort | high effort |
GDPval-AA v2.1 (Elo) Real-world work tasks; a rating against other models, not a percentage | 1624, source: Artificial Analysis, high effort | high effort |
LCR Long-context reasoning, AA-LCR v1.1 | 79.7%, source: Artificial Analysis, high effort | high effort |
Omni Accuracy Omniscience knowledge accuracy | 49.9%, source: Artificial Analysis, high effort | high effort |
Non-Hallucination 1 − hallucination rate | 84.9%, source: Artificial Analysis, high effort | high effort |
More from Google
See all 16 Google models- Gemini 2.5 Profrom $1.25 in · 1.05M · GPQA 84.4% · 3 providers
- Gemini 3.1 Flash-Litefrom $0.25 in · 1.05M · GPQA 82.2% · 3 providers
- Gemini 3.1 Pro Previewfrom $2.00 in · 1.05M · GPQA 94.1% · 3 providers
- Gemini 3.5 Flashfrom $1.50 in · 1.05M · GPQA 92.2% · 3 providers
- Gemini 3.5 Flash-Litefrom $0.30 in · 1.05M · GPQA 83.8% · 2 providers
- Gemini 3.6 Flashfrom $1.50 in · 1.05M · GPQA 92.8% · 2 providers
- Gemini 3.7 Flashfrom $0.75 in · 1.05M · GPQA 94.5% · 2 providers
- Gemini 3.8 Flashfrom $1.50 in · 1.05M · GPQA 95.3% · 2 providers
- Gemini 3 Deep ThinkMMMU-Pro (standard) 81.5%
- Gemma 3 270M32k · GPQA 22.4%
- Gemma 4 12B256k · GPQA 75.3%
- Gemma 4 26B A4B256k · GPQA 79.2%