One-line summary
This dataset covers key facts on Gemini 4 Argon released by Google on October 13, 2026: introductory API pricing of $2 per million input tokens and $10 per million output tokens (until 2026-11-30, 95% off cached input, then $4/$20), rank #1 on Vals Index, and 87% on Embodied Reasoning (ERQA).
Data Notes
Sources: Google official blog (published October 13, 2026), Google DeepMind model page and Google AI for Developers pricing page; figures follow official disclosures as of October 13, 2026. Prices in USD per million tokens (M tokens).
Core Data
Table 1: API Pricing
| field | description | value | notes/validity |
|---|---|---|---|
| model_name | Model name | Gemini 4 Argon | Flagship model, globally available 2026-10-13 |
| price_per_million_input_usd | Introductory input price (USD/M tokens) | 2 | Introductory, until 2026-11-30 |
| price_per_million_output_usd | Introductory output price (USD/M tokens) | 10 | Introductory, until 2026-11-30 |
| cached_input_discount | Cached input discount | 95% | Cached input tokens billed at 95% off input price |
| standard_price_per_million_input_usd | Standard input price (USD/M tokens) | 4 | Effective 2026-12-01 |
| standard_price_per_million_output_usd | Standard output price (USD/M tokens) | 20 | Effective 2026-12-01 |
| availability | Available channels | AI Studio, Gemini platform API, Antigravity, Gemini app & AI Mode in Search | Subscriptions: AI Ultra for Antigravity/Gemini app; AI Ultra/Pro for AI Mode in Search |
Table 2: Benchmarks and Internal Results
| field | description | value | notes/date |
|---|---|---|---|
| vals_index_rank | Vals Index rank | 1 | Overall benchmark, announced 2026-10-13 |
| erqa_score_pct | Embodied Reasoning (ERQA) score | 87% | Announced 2026-10-13 |
| agentic_code_submissions_change | Agentic code submissions change | +35% (more than) | Google internal & early customer use |
| code_rollback_change | Code rollback change | Down by nearly a third | Google internal use |
| gemma_e2b_throughput_before | Gemma 4 E2B throughput (A100, before) | ~37,900 tokens/s | End-to-end inference |
| gemma_e2b_throughput_after | Gemma 4 E2B throughput (A100, after) | Over 108,000 tokens/s | 2.85x speedup |
| chrome_legacy_code_refactor | Chrome legacy code refactor scale | ~6 million lines | Within Chromium's ~40 million-line codebase |
Table 3: Related Programs
| field | description | value | notes/date |
|---|---|---|---|
| scientist_program_quota | Free program for scientists | 10,000 seats | 6 months of top-tier Antigravity; onboarded on a rolling basis subject to eligibility, capacity and selection |
| fairwind_program | Fairwind program expansion | Hundreds of partners | Per Google |
| wiz_scan_for_good | Wiz security service | Government data & healthcare infrastructure | Per Google's "Scan for Good" |
Field Definitions
- price_per_million_input_usd / output_usd: input/output price per million tokens (USD); the introductory price applies until 2026-11-30 (official timezone not disclosed), then the standard price takes effect.
- cached_input_discount: billing discount on cached input tokens in Context Caching scenarios.
- vals_index_rank / erqa_score_pct: benchmark results disclosed by Google; evaluation methodology per official notes.
- Table 2 internal figures are Google-disclosed results from "internal and early customer use", not independent third-party evaluations.