# Deepseek-V4-Flash

2026-09-20 · https://omnimodel.me/catalog/deepseek-v4-flash

Model ID: `deepseek-v4-flash`

## Capabilities

Text

## Pricing

Effective since 2026-09-24

| Item | Price |
| --- | --- |
| Input / 1M tokens | **0.201 USD** · Official 0.3 USD · Save 33% vs official |
| Output / 1M tokens | **0.804 USD** · Official 1.2 USD · Save 33% vs official |
| Cache read / 1M tokens | **0.004 USD** · Official 0.006 USD · Save 33% vs official |

## Reasoning effort

The platform handles reasoning in two steps: it turns each client's setting into one level, then sends a level the upstream accepts. A level the upstream rejects becomes the next higher accepted level, or its highest one; rewrites show up in usage records.

### Client setting → platform level

| Client and API | Sent as | Platform level |
| --- | --- | --- |
| Codex · Responses | `reasoning.effort` | same value |
| OpenCode / pi / Hermes · Chat | `reasoning_effort` | same value |
| Claude Code · Anthropic | `thinking.type=enabled, budget_tokens 1–512` | minimal |
| Claude Code · Anthropic | `budget_tokens 513–1024` | low |
| Claude Code · Anthropic | `budget_tokens 1025–8192` | medium |
| Claude Code · Anthropic | `budget_tokens 8193–24576` | high |
| Claude Code · Anthropic | `budget_tokens above 24576` | xhigh |
| Claude Code · Anthropic | `thinking.type=enabled without budget_tokens` | auto |
| Claude Code · Anthropic | `thinking.type=adaptive with output_config.effort` | the output_config.effort value |
| Claude Code · Anthropic | `thinking.type=adaptive without an effort` | xhigh |
| Claude Code · Anthropic | `thinking.type=disabled` | none |
| Gemini · generateContent | `thinkingConfig.thinkingLevel` | same value |
| Gemini · generateContent | `thinkingConfig.thinkingBudget` | budget ranges above; -1 is auto, 0 is none |

### Platform level → level sent upstream

| Platform level | Sent upstream |
| --- | --- |
| `none` | `none` |
| `minimal` | `minimal` |
| `low` | `low` |
| `medium` | `medium` |
| `high` | `high` |
| `xhigh` | `xhigh` |
| `max` | `max` |
| `auto` | `auto` |

## Usage

OpenAI-compatible base URL `https://omnimodel.me/v1`, set model to `deepseek-v4-flash`.