GPT Image 2.5 is live — OpenAI's newest image model, targeted edits that leave the rest of the frame alone

Gemini 4 Neon — Frontier Reasoning for Complex Workflows

Use Gemini 4 Neon for multi-file software engineering, long tool-heavy agents, science and knowledge work. Combine a unified 1M context window with four thinking levels.

Multimodal inputFunction callingStructured outputStreaming
Highlights
1M shared context · four thinking levels · 64K hops
Best for
Complex software engineering, long tool agents and science
Input
Text, images, video, audio and PDF
Output
Text · up to 65,536 tokens per hop
Gemini 4 Neon

Ask Gemini 4 Neon

Start with the prompt in the composer or choose an example.

Enter to send · Shift+Enter for a new line

Pricing

Credit-based — 1 credit = $0.001 USD. Pay only for completed generations.

Pricing by group · gemini-4-neon

Loading groups…

USD per 1M tokens. Tool charges may apply.

What is Gemini 4 Neon?

Gemini 4 Neon is Google's stable frontier reasoning model released on November 18, 2026, adapted from Gemini 4 Carbon to sit between Gemini 4 Hydrogen and Carbon. It takes text, images, video, audio and PDF inputs and outputs text within one shared 1,048,576-token context window. Google built Gemini 4 Neon for high-frequency complex tasks, including repository-level debugging, multi-step tool agents, quantitative knowledge work and code hardening, where Hydrogen lacks depth and Carbon is heavier than needed. Test it on a verifiable task, such as patching a race condition with a failing test or synthesizing a dense spreadsheet and PDF packet.

What you can build with this model

Real-world workflows and production use cases you can build and ship with this model.

Gemini 4 Neon software engineering concept showing linked editor panes, a diff thread and a verification tile.

Debug complex codebases faster with Gemini 4 Neon

Give Gemini 4 Neon the failing test, related modules and constraints for a multi-file refactor or root-cause security patch. Google designed it to finish repository-level edits cleanly in fewer steps while reducing shortcuts and unverified claims. Keep repository conventions in the prompt, then run your test suite locally to confirm every diff.

Gemini 4 Neon agent workflow concept with chained tool blocks connected to a central glass hub.

Run long agentic workflows with Gemini 4 Neon

Put Gemini 4 Neon behind tool-heavy sessions that span dozens of reads, searches and code checks. It handles context compaction across long runs and cuts redundant file reads, working standalone or as an orchestrator paired with Hydrogen workers. Your code still executes every function call and enforces permissions before any action runs.

Gemini 4 Neon multimodal analysis concept with stacked documents, a grid tile and a clean summary slab.

Synthesize research and data with Gemini 4 Neon

Feed Gemini 4 Neon dense PDFs, spreadsheets, charts, audio or video across its unified 1M context window to produce structured legal, financial or scientific deliverables. It supports native JSON schemas and Google Search or Maps grounding when configured. Ask it to cite exact passages and check every figure against the source file.

Why teams choose Gemini 4 Neon

Gemini 4 Neon balances depth and speed

Gemini 4 Neon fills the middle tier of the Gemini 4 lineup: deeper reasoning than Hydrogen with leaner token use and faster end-to-end completion than Carbon. Teams can standardize high-frequency engineering and analytical pipelines on Gemini 4 Neon, reserving Carbon for the hardest long-horizon problems.

Gemini 4 Neon four thinking levels

Gemini 4 Neon defaults to high and lets you tune down to low or medium for responsive tools, or up to xhigh for hard engineering and scientific constraints. Google advises evaluating all four Gemini 4 Neon levels with generous output room, then picking the lowest setting that clears your quality bar.

Gemini 4 Neon native protocol parity

Gemini 4 Neon shares the exact GenerateContent request contract used by Carbon, Argon and Hydrogen. Existing Gemini 4 clients switch by changing the model ID and checking the default thinking level, while keeping function calling, thought signatures and explicit continuation intact.

Start testing Gemini 4 Neon

  1. step 01

    Select a Gemini 4 Neon group

    Open the Gemini 4 Neon Playground with your reAPI account, compare the discovered group prices and pick a route. Start with a concrete multi-file bug, tool sequence or document synthesis task you can grade objectively.

    Open
  2. step 02

    Tune Gemini 4 Neon thinking levels

    Run your benchmark prompt at Gemini 4 Neon's default high level first, then test medium and xhigh on the same input. Compare answer accuracy, tool-call count and total billed tokens before locking a production setting.

    Open
  3. step 03

    Validate Gemini 4 Neon outputs

    Execute generated tests, check structured JSON against your schema and audit tool arguments before committing changes. Gemini 4 Neon can still hallucinate or miss edge cases, so keep automated checks and human review in the loop.

    Open

Capabilities

Gemini 4 Neon multimodal inputs

Gemini 4 Neon reads text, images, video, audio and PDF inputs and generates text output. Image generation, audio generation and the Live API are not supported on Gemini 4 Neon; use dedicated media models for generated assets.

Gemini 4 Neon unified 1M context

Gemini 4 Neon shares 1,048,576 tokens across prompt input, thinking tokens, tool calls and final text. Plan prompt size with room for reasoning and output, because these budgets come from one combined pool.

Gemini 4 Neon thinking controls

Gemini 4 Neon supports low, medium, high (default) and xhigh thinking levels. Thinking cannot be disabled, and native minimal requests are rejected. Request thought summaries separately with includeThoughts when needed.

Gemini 4 Neon tools and structured output

Google documents native function calling, code execution, computer use, file search, URL context, Search and Maps grounding and JSON Schema structured outputs for Gemini 4 Neon. Your application runs client functions and validates returned JSON.

Gemini 4 Neon 64K hops and long decode

Gemini 4 Neon returns up to 65,536 output tokens per single JSON or streaming request. Longer outputs use explicit continuationToken loops within the remaining shared 1M context window, with each hop billed separately.

Gemini 4 Neon implicit caching

Gemini 4 Neon uses implicit caching automatically on eligible prompts of at least 4,096 tokens, discounting cache hits when reported in usage. Explicit cached-content resources are unsupported on Gemini 4 Neon.

Gemini 4 Neon vs Gemini 4 Carbon

Google adapts Gemini 4 Neon from Gemini 4 Carbon to deliver frontier reasoning with faster completion on high-frequency complex workflows. The table below compares documented specifications rather than an independent reAPI benchmark; Gemini 4 Hydrogen remains the lighter choice for fast everyday execution.

Decision point
Gemini 4 Neon on reAPI
Gemini 4 Carbon
Intended workload
High-frequency complex coding, long agents, knowledge work and science
Hardest long-horizon architecture, debugging, science and security
Thinking levels
low / medium / high / xhigh
low / medium / high
Default thinking
high
medium
Unified context
1,048,576 tokens, shared
1,048,576 tokens, shared
Non-streaming REST hop
Up to 65,536 output tokens
Up to 32,768 output tokens
Caching and pricing tier
Implicit caching only; flat paid-tier token rates across 1M window
Implicit caching only; flat paid-tier token rates across 1M window

Sources: Google's Gemini 4 Neon model documentation, migration guide and DeepMind model card, checked November 20, 2026. Actual quality, speed and total token spend depend on prompt and thinking level.

Frequently asked questions

Common questions about this model.

Google built Gemini 4 Neon for high-frequency complex software engineering, long-running tool agents, analytical knowledge work, science and code security hardening. Adapted from Gemini 4 Carbon, Gemini 4 Neon sits between Hydrogen and Carbon so teams get frontier reasoning with faster end-to-end completion. Always verify Gemini 4 Neon on tasks with clear acceptance checks.

docs/api/gemini-4-neon

API reference

Drop-in code and the full parameter table.

curl https://reapi.ai/api/v1beta/models/gemini-4-neon:generateContent \
  -H "x-goog-api-key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"contents":[{"role":"user","parts":[{"text":"Audit this retry queue for race conditions and outline a minimal fix."}]}],"generationConfig":{"maxOutputTokens":4096,"thinkingConfig":{"thinkingLevel":"high"}}}'

Parameter validation failed

Rejected synchronously with the failing field named. Check enum values, ranges, and URL fields — nothing is charged.

Authentication or balance issue

401 means a missing or invalid Bearer key; 402 means the reserve exceeds your balance. Manage keys and credits in the console.

Content or material rejected

Reference material and outputs pass automated moderation. Rejected tasks fail with a clear error and a full refund.

Task failed or timed out

A task that reaches a failed state is never charged — the reserve refunds automatically. Keep the task id and retry when ready.

start building

Test Gemini 4 Neon on a real workflow

Check the live Gemini 4 Neon group rates, pick a thinking level and send a native GenerateContent request with your reAPI key. Read the API docs for streaming, tool turns and continuation.