New Flagship Models Added: DeepSeek-V4-Flash and DeepSeek-V4-Pro

Date: 2026-04-24 / (1405-02-04)

Summary

We announce the addition of DeepSeek's new flagship models, deepseek-v4-flash and deepseek-v4-pro, on the v1/chat/completions endpoint. DeepSeek V4 introduces a 1M context window as standard, novel attention with token-wise compression and DeepSeek Sparse Attention (DSA), and dual thinking/non-thinking modes. The legacy deepseek-chat and deepseek-reasoner aliases are now routed to the V4 family and will be deprecated on July 24, 2026.

---

Details

DeepSeek

DeepSeek-V4-Flash

We introduce DeepSeek-V4-Flash (deepseek-v4-flash), a fast, efficient, and economical flagship with 284B total / 13B active parameters. Reasoning capabilities closely approach V4-Pro and performance is on par with V4-Pro on simple agent tasks, while offering smaller parameter size, faster response times, and highly cost-effective pricing. Documentation

DeepSeek-V4-Pro

We introduce DeepSeek-V4-Pro (deepseek-v4-pro), DeepSeek's most capable flagship with 1.6T total / 49B active parameters. Per DeepSeek's benchmarks, V4-Pro delivers open-source SOTA in Agentic Coding benchmarks, world-class reasoning that rivals top closed-source models in Math/STEM/Coding, and rich world knowledge that trails only Gemini-3.1-Pro among current open models. Documentation

Key Features (both models):

• Context Window: 1M tokens (default across all DeepSeek V4 services)
• Max Output Tokens: Up to 384K
• Dual Modes: Supports both thinking (default) and non-thinking modes — toggle via extra_body={"thinking": {"type": "enabled"/"disabled"}}
• Thinking Effort Control: reasoning_effort: "high"/"max" (low/medium are mapped to high, xhigh is mapped to max)
• Structural Innovation: Token-wise compression + DeepSeek Sparse Attention (DSA) for peak long-context efficiency
• Advanced Capabilities: JSON Output, Tool Calls, Chat Prefix Completion (Beta)
• FIM Completion (Beta): Available in non-thinking mode only
• Agent Integration: Seamlessly integrated with leading AI agents like Claude Code, OpenClaw, and OpenCode
• Endpoint Support: Available on v1/chat/completions

Pricing Details:

Model	Input (Cache Miss)	Input (Cache Hit)	Output
deepseek-v4-flash	$0.14 / 1M tokens	$0.028 / 1M tokens	$0.28 / 1M tokens
deepseek-v4-pro	$1.74 / 1M tokens	$0.145 / 1M tokens	$3.48 / 1M tokens

Model Routing and Legacy Aliases

To preserve backward compatibility, AvalAI now routes the legacy aliases to the new V4 flagships:

• deepseek-chat → routed to deepseek-v4-flash (non-thinking mode)
• deepseek-reasoner → routed to deepseek-v4-pro (thinking mode)

Pricing impact for legacy aliases:

• deepseek-chat pricing is unchanged — same as deepseek-v4-flash ($0.14 / $0.028 / $0.28 per 1M tokens for input / cached input / output).
• deepseek-reasoner pricing has been raised to match deepseek-v4-pro ($1.74 / $0.145 / $3.48 per 1M tokens for input / cached input / output) to reflect the upgrade in underlying capability.

⚠️ Deprecation Notice

Per DeepSeek's official announcement, deepseek-chat and deepseek-reasoner will be fully retired and inaccessible after July 24, 2026, 15:59 (UTC). Users are encouraged to migrate to the explicit deepseek-v4-flash and deepseek-v4-pro model names. See the Deprecations page for full details.

API Request/Response Examples

Example Request — DeepSeek-V4-Flash (non-thinking mode)

curl https://api.avalai.ir/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "deepseek-v4-flash",
    "messages": [
      {
        "role": "user",
        "content": "Explain the key benefits of DeepSeek Sparse Attention (DSA) in one paragraph."
      }
    ]
  }'

Example Response

{
  "id": "chatcmpl-deepseek-v4-flash-xxxxx",
  "created": 1745481600,
  "model": "deepseek-v4-flash",
  "object": "chat.completion",
  "system_fingerprint": null,
  "choices": [
    {
      "finish_reason": "stop",
      "index": 0,
      "message": {
        "content": "DeepSeek Sparse Attention (DSA) combines token-wise compression with a sparse attention pattern to dramatically reduce compute and memory cost in long-context scenarios. This enables 1M-token default context across DeepSeek services while keeping inference fast and economical.",
        "role": "assistant",
        "reasoning_content": null,
        "thinking_blocks": [],
        "annotations": []
      }
    }
  ],
  "usage": {
    "completion_tokens": 58,
    "prompt_tokens": 22,
    "total_tokens": 80,
    "completion_tokens_details": null,
    "prompt_tokens_details": {
      "audio_tokens": null,
      "cached_tokens": 0,
      "text_tokens": 22,
      "image_tokens": null
    }
  },
  "estimated_cost": {
    "unit": "0.0000193200",
    "irt": 2.21,
    "exchange_rate": 114600
  }
}

Example Request — DeepSeek-V4-Pro (thinking mode)

curl https://api.avalai.ir/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "deepseek-v4-pro",
    "messages": [
      {
        "role": "user",
        "content": "Design a fault-tolerant architecture for a global payment processor that handles 50,000 TPS."
      }
    ],
    "reasoning_effort": "high"
  }'

Example Response (thinking mode)

{
  "id": "chatcmpl-deepseek-v4-pro-xxxxx",
  "created": 1745481700,
  "model": "deepseek-v4-pro",
  "object": "chat.completion",
  "system_fingerprint": null,
  "choices": [
    {
      "finish_reason": "stop",
      "index": 0,
      "message": {
        "content": "Here is a fault-tolerant architecture for a global payment processor ...",
        "reasoning_content": "The user wants a fault-tolerant architecture for 50K TPS globally. I should cover regional active-active deployment, consensus for ledger consistency, idempotency keys, saga-based orchestration, and disaster recovery ...",
        "role": "assistant",
        "thinking_blocks": [],
        "annotations": []
      }
    }
  ],
  "usage": {
    "completion_tokens": 1842,
    "prompt_tokens": 35,
    "total_tokens": 1877,
    "completion_tokens_details": null,
    "prompt_tokens_details": {
      "audio_tokens": null,
      "cached_tokens": 0,
      "text_tokens": 35,
      "image_tokens": null
    }
  },
  "estimated_cost": {
    "unit": "0.0064703100",
    "irt": 741.5,
    "exchange_rate": 114600
  }
}

SDK Usage Examples

Bash

curl https://api.avalai.ir/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "deepseek-v4-flash",
    "messages": [
      {
        "role": "user",
        "content": "Summarize the advantages of a 1M token context window for document analysis."
      }
    ]
  }'

Python

from openai import OpenAI

client = OpenAI(api_key="your-avalai-api-key", base_url="https://api.avalai.ir/v1")

# Non-thinking mode (fast, economical)
response = client.chat.completions.create(
    model="deepseek-v4-flash",
    messages=[
        {
            "role": "user",
            "content": "Summarize the advantages of a 1M token context window for document analysis.",
        }
    ],
    extra_body={"thinking": {"type": "disabled"}},
)

print(response.choices[0].message.content)

# Thinking mode (deep reasoning)
reasoning_response = client.chat.completions.create(
    model="deepseek-v4-pro",
    messages=[
        {
            "role": "user",
            "content": "Prove that the sum of the first n odd numbers equals n^2.",
        }
    ],
    reasoning_effort="high",
    extra_body={"thinking": {"type": "enabled"}},
)

print(reasoning_response.choices[0].message.reasoning_content)
print(reasoning_response.choices[0].message.content)

JavaScript

import { OpenAI } from "openai";

const client = new OpenAI({
  apiKey: process.env.AVALAI_API_KEY,
  baseURL: "https://api.avalai.ir/v1",
});

// Non-thinking mode (fast, economical)
const response = await client.chat.completions.create({
  model: "deepseek-v4-flash",
  messages: [
    {
      role: "user",
      content: "Summarize the advantages of a 1M token context window for document analysis.",
    },
  ],
});

console.log(response.choices[0].message.content);

// Thinking mode (deep reasoning)
const reasoningResponse = await client.chat.completions.create({
  model: "deepseek-v4-pro",
  messages: [
    {
      role: "user",
      content: "Prove that the sum of the first n odd numbers equals n^2.",
    },
  ],
  reasoning_effort: "high",
});

console.log(reasoningResponse.choices[0].message.reasoning_content);
console.log(reasoningResponse.choices[0].message.content);

Migration Guidance

• No action required for existing integrations: calls to deepseek-chat and deepseek-reasoner continue to work and are automatically routed to the V4 family until July 24, 2026.
• Recommended: update your model identifiers to deepseek-v4-flash or deepseek-v4-pro to benefit from the 1M context, explicit mode control via the thinking parameter, and to avoid disruption when the legacy aliases are retired.
• Pricing awareness: if you currently use deepseek-reasoner, note that its output pricing now reflects deepseek-v4-pro ($3.48 / 1M tokens output) to match the upgraded underlying model.

---

Related Links

• DeepSeek Models Documentation
• Deprecations
• Reasoning Models Guide
• Chat Completions API Reference
• Pricing Information