Gemini 4 Neon
1M · $1.50/1M tokens
gemini-4-neonTurn code, documents and images into patches, evidence-linked analysis and plans your agent can act on. Gemini 4 Argon sustains multi-step reasoning for software engineering and professional knowledge work, with streaming and tool calling through BeatAPI.
set up https://beatapi.io/SKILL.mdGive this to your agent
Chat with Gemini 4 Argon here. Every message is a real request, billed to your balance at the same price as the API.
Input, cache-hit input and output are billed separately per million tokens. Repeated repository instructions or document prefixes cost less when the response reports a cache hit; a repeated prompt alone does not guarantee one.
Current rates apply to the introductory period. Its end date is not fixed. Set an output budget for each task and use returned usage to track actual cost, including reasoning tokens counted as output.
| Model | Specifications and pricing | beatapi |
|---|---|---|
| Gemini 4 Argon | Gemini 4 ArgonInput · per 1M tokens | $1.00 / 1M tokens |
| Gemini 4 Argon | Gemini 4 ArgonCache-hit input · per 1M tokens | $0.05 / 1M tokens |
| Gemini 4 Argon | Gemini 4 ArgonOutput · per 1M tokens | $5.00 / 1M tokens |
Compare context specifications and published pricing, with links to model details and API documentation.
1M · $1.50/1M tokens
gemini-4-neon· $1.00/1M tokens
gemini-4-carbon· $0.75/1M tokens
gemini-4-hydrogen· $1.00/1M tokens
gemini-4-argon1M · $1.00/1M tokens
gemini-3.1-pro-preview1M · $0.375/1M tokens
gemini-3.8-flash1M · $0.375/1M tokens
gemini-3.7-flash1M · $0.375/1M tokens
gemini-3.6-flashChoose by task length, visual material and cost. Argon provides a million-token output budget; Pro and Flash's million-token input limits and 65,536-token output limits are listed separately.
| Comparison | Gemini 4 Argon | Gemini 3.1 Pro | Gemini 3.8 Flash | Gemini 3.7 Flash | Gemini 3.6 Flash |
|---|---|---|---|---|---|
| Task fit and selection | Complex long-horizon engineering, legal and financial knowledge work, and cyber defense. Choose for sustained deep reasoning, professional charts, long-video details and larger output headroom. | Complex reasoning, software engineering and reliable multi-step tool execution. Choose for analysis over million-token inputs with outputs within 65,536 tokens. | Long-horizon software engineering, autonomous agents and complex enterprise workflows. Choose for multimodal understanding with attention to response speed, request frequency and cost. | Previous-generation Flash. Google API has deprecated it and redirects requests to 3.8 Flash. Prefer 3.8 Flash for new integrations; validate the actual service route when migrating an existing app. | Code generation, agent execution and spatial reasoning. Choose for rapid coding iterations, repeated tool loops and applications retaining an existing 3.6 Flash integration. |
| Inputs | Text · images · video | Text · images · video · audio · PDF | Text · images · video · audio · PDF | Text · images · video · audio · PDF | Text · images · video · audio · PDF |
| Model input token limit | — | 1,048,576 | 1,048,576 | 1,048,576 | 1,048,576 |
| Model output token limit | 1,000,000 | 65,536 | 65,536 | 65,536 | 65,536 |
| Input · per 1M tokens | $1.00 | $1.00 | $0.375 | $0.375 | $0.375 |
| Cache-hit input · per 1M tokens | $0.05 | $0.10 | $0.0375 | $0.0375 | $0.0375 |
| Output · per 1M tokens | $5.00 | $6.00 | $1.875 | $1.875 | $1.875 |
| API | gemini-4-argon | gemini-3.1-pro-preview | gemini-3.8-flash | gemini-3.7-flash | gemini-3.6-flash |
Input and output specifications are separate. Argon's announced 1M figure is an output limit, not a derived input value. Modalities describe model capabilities; prices are BeatAPI retail rates per million tokens. For 3.1 Pro inputs above 200K tokens, input and output cost $2 and $9 respectively; cache hits are billed separately. The 3.7 Flash redirect refers to Google API, not inferred routing on another service.
Gemini 4 Argon combines coding, reasoning and multimodal understanding to sustain long, multi-step tasks: tracing failures, reconciling documents and visual evidence, and deciding what to change.
Combine an issue, relevant source files and failing tests to trace dependencies and develop a patch. Argon can reason through migration steps and implementation trade-offs; your agent supplies repository access, runs tests and applies approved changes.
Compare clauses, research notes and tabular text to explain agreements, conflicts and missing evidence. Request section references alongside conclusions so a reviewer can trace each claim back to the supplied material.
Gemini 4 Argon analyzes professional charts, identifies details in long videos and takes action across a series of documents. Chart changes, video events and documentary evidence can inform the same multi-step analysis.
Define functions for repository lookup, document retrieval or internal records. Argon returns a function name and arguments; your application executes it and returns the result for the next reasoning step. External access and write permissions stay under your control.
| Benchmark | Gemini 4 Argon | Gemini 4 Carbon | GPT-6 Astra | Claude Opus 5.5 | Claude Fable 5.1 |
|---|---|---|---|---|---|
| AA Index v5.0 | 50 | 59 | 57 | 58 | 49 |
| Vals Index v2.2 | 69.2% | 70.4% | 65.0% | 70.2% | — |
| SWE-sweep | 4.3% | 9.6% | 9.5% | 3.6% | 2.9% |
| FrontierSWE 2 | 55.0% | 61.2% | 65.5% | 62.3% | 56.3% |
| Terminal-Bench 4.0 (AA) | 57.1% | 65.2% | 59.1% | 59.6% | 52.0% |
| Agents' Last Exam v1.1 (Score) | 69.0% | 77.4% | 73.1% | 71.6% | — |
| OSWorld-2.1 (Offline subset) | 74.7% | 81.5% | 80.9% | 87.2% | 76.8% |
| AutomationBench (Zapier) | 51.3% | 53.9% | 41.4% | 42.5% | 31.4% |
| Vending-Bench 2 (Bank balance) | $13,718 | $16,484 | $15,515 | $9,235 | $5,422 |
| Chartography | 71.6% | 72.5% | 71.0% | 66.3% | 46.2% |
| Blueprint-Bench 2 (Similarity score) | 54.4% | 58.6% | 49.7% | 51.2% | 41.9% |
| GraphWalks (256k to 1M, BFS F1) | 84.2% | 96.2% | 71.8% | 66.8% | 65.0% |
| RiemannBench | 76.0% | 80.0% | 72.0% | 69.6% | 65.6% |
| Terminal-Bench-Science 0.1 (AA) | 52.9% | 64.8% | 63.3% | 59.0% | 43.3% |
| CWE-bench | 68.0% | 71.0% | 68.0% | 67.0% | 58.0% |
| ExploitGym | 42.0% | 54.8% | 42.4% | 34.5% | — |
| SEC-bench Pro | 83.4% | 88.2% | 85.4% | — | — |
An old adapter, a new library specification and existing tests form a migration brief. Produce an interface mapping, a sequenced patch plan and regression cases for the changed behavior, ready for a pull-request review.
Two contract versions and a review policy become a clause-level change register: altered obligations, conflicting definitions, unresolved references and draft questions for counsel. The result supports legal review rather than replacing it.
Analyze a financial chart alongside extracted figures and reporting notes to explain changes, with units, periods and source references. Keep supplied values separate from assumptions so an analyst can check the explanation before using it in a report.
Logs, a deployment diff and a topology screenshot become a timeline, ranked failure hypotheses and a targeted test plan. Repository and monitoring tools can supply additional evidence; proposed commands remain subject to your execution policy.
Compare published specifications and evaluate output quality on your own workload.
| Comparison | Gemini 4 Argon | GPT-6.2 Sol | Claude Sonnet 5.5 | Grok 4.7 |
|---|---|---|---|---|
| Context window | — | 1.05M | 1M | 500K |
| Inputs | Text · Image · Video | Text · Image | Text · Image | Text |
| Price from | $1.00/1M tokens | $0.80/1M tokens | $1.00/1M tokens | $0.70/1M tokens |
| API | Gemini 4 Argon | GPT-6.2 Sol | Claude Sonnet 5.5 | Grok 4.7 |
Call gemini-4-argon with OpenAI Chat Completions, Responses, Anthropic Messages or native Gemini generateContent. Keep the request and response shape of your chosen protocol; use native Gemini for Gemini-specific fields.
Sign up or sign in to open the Dashboard. Check your balance and open the credits page if you need to add funds.
Open DashboardBring sustained reasoning to work that combines engineering detail, evidence and application actions.
A model output ceiling of one million tokens leaves room for extended reasoning and substantial deliverables. Use a smaller task-specific cap for routine work; the ceiling is not a target or a promised answer length.
Code snippets, chart images and written constraints can inform the same analysis. This is especially useful when a diagram explains a dependency that text alone would miss.
Stable repository rules and shared document prefixes recur across agent turns. Cache-hit input has its own lower rate, while input and output remain separately visible in usage for budgeting.
OpenAI-compatible clients can use Chat Completions or Responses; Anthropic SDK clients can use Messages; Gemini applications can keep native contents and parts. One BeatAPI key also lets the application call a separate model for shorter tasks without changing its account integration.
Model identity, input support, token costs and integration details.
Gemini 4 Argon is Google's Gemini 4 model for complex software engineering, legal, financial and defensive cybersecurity work. It combines coding, sustained reasoning and multimodal understanding to analyze text, images and long video across long, multi-step tasks.
Use it for code changes with several dependencies, document comparison, chart-supported analysis and investigations that need successive tool results. Supply relevant material and define the deliverable: a patch, clause register, variance explanation or test plan. The application still runs tools, tests and approval steps.
Yes. Use model ID gemini-4-argon with your BeatAPI key through Chat Completions, Responses, Messages or native Gemini requests, with streaming and tool calling.
The price table separates input, cache-hit input and output per million tokens. Thinking counted in output usage is billed as output even when it is not part of the visible answer. Cached pricing applies only to reported cache-hit tokens. The current rates are introductory; no end date is fixed.
Gemini 4 Argon has a one-million-token output limit, expanded from the previous 64K. This headroom supports sustained deep reasoning and hundreds of thousands of tokens in a single trajectory for longer, more complex tasks.
Use the exact identity Gemini 4 Argon and request ID gemini-4-argon. Do not substitute a predicted Gemini 4 Pro name or undocumented low, medium or high model IDs. Gemini 3.8 Flash and 3.1 Pro are separate models with their own prices and limits.
Yes. Gemini 4 Argon's multimodal understanding supports professional chart analysis, identifying details in long videos and knowledge work across multiple documents. Understanding existing images and video is distinct from generating new media.
No. The model chooses from the functions you define and returns arguments. Your application validates them, executes the function and sends its result back. Restrict accessible resources and require approval for writes. Calling the model does not grant it access to external systems.
Keep Chat Completions or Responses if your application already uses an OpenAI-compatible client, or Messages if it uses an Anthropic SDK. Use native Gemini for contents, parts and Gemini-specific fields. These formats have different request and response structures; consult the API tab rather than copying fields between protocols.
Argon is not in the signup-credit model allowlist. Fund your account before calling it. It is distinct from the separately listed free models, which have their own availability and rate limits.
Connect Gemini 4 Argon with other production-ready APIs in the BeatAPI catalog.
Guides and practical notes to help you get more from Gemini 4 Argon.