DeepSeek V4 Pro Launches: Higher Specs, Same Price, Big Agent Play

DeepSeek's V4 Pro Is Now Official

DeepSeek just dropped the production version of its V4 Pro model. The updated model page shows the version tag as DeepSeek-V4-Pro-0813. If you've been waiting for a more powerful option in the V4 lineup, this is it.

The headline specs are hard to ignore: a 1 million token context window and a maximum output of 384,000 tokens. That's a lot of room for long codebases, hefty document dumps, or agent workflows that need to keep state across many steps. For developers who've been hitting context limits, this removes a major bottleneck.

Both thinking and non-thinking modes are supported, with thinking mode enabled by default. That's a nice touch for tasks that benefit from reasoning but also gives you the option to cut latency when you don't need it.

Pricing: The Same, For Now

Here's the twist: DeepSeek had announced plans to raise API prices across the board, but V4 Pro's current rates are unchanged. That means you're getting the new model at the same cost as before—at least for now.

Per million tokens, V4 Pro charges:

• 0.025 yuan for cached input
• 3 yuan for uncached input
• 6 yuan for output

Compare that to V4 Flash, which runs 0.02 yuan (cached), 1 yuan (uncached), and 2 yuan (output). So Pro costs three times as much on the main input and output lines. But the absolute numbers are still low enough that even heavy use won't break the bank.

Concurrency: Pro Is Not the Speed Demon

If you're thinking about scaling, note the concurrency limits. Flash allows 2,500 concurrent requests, while Pro is capped at 500. That's a clear signal: Flash is built for high-frequency, high-volume calls, whereas Pro is more about raw capability per request.

For most agent or coding tasks, 500 concurrent requests is plenty. But if you're running a massive parallel pipeline, you might want to mix in Flash for the bulk work and reserve Pro for the tricky parts.

Agent-Ready Features

DeepSeek didn't just bump up the specs; they added features that matter for agent development. V4 Pro supports JSON Output, Tool Calls, the Responses API, and even the Anthropic API format. That last one is interesting—it means you can plug it into tools built for Claude's API with minimal changes.

There's also beta support for conversation prefix continuation and FIM (fill-in-the-middle) completion. FIM is limited to non-thinking mode, which makes sense because quick autocomplete doesn't need deep reasoning.

The API also accepts both OpenAI and Anthropic request formats, so you can switch clients without rewriting everything. That's a practical move for teams juggling multiple providers.

Flash vs. Pro: Which One Should You Use?

The V4 family now has a clear split. Flash is the budget workhorse—low cost, high concurrency, and perfect for repetitive tasks like batch processing or simple classification. Pro is the brainier sibling: better for complex reasoning, long-context analysis, and agent loops that need tool calls and structured output.

In practice, you might start with Flash for prototyping, then move to Pro when you hit accuracy or context limits. Or run both in parallel: Flash for pre-processing, Pro for the final decision step.

Why the Anthropic API Compatibility Matters

Supporting the Anthropic API format is a bigger deal than it might seem. Many agent frameworks and coding assistants are built around Anthropic's API conventions. By speaking that language, DeepSeek makes it trivial to swap in V4 Pro as a backend. You don't need to fork your codebase or write a translation layer.

It also hints at where DeepSeek is heading: they want to be a drop-in replacement for existing AI services, not just another API with its own quirks. That's a smart strategy for winning over developers who are tired of vendor lock-in.

What About the Price Hike?

DeepSeek's earlier announcement about raising prices might have made some people nervous. But for now, the rates are stable. That could change soon, though, so if you're planning to build on V4 Pro, it might be worth locking in your usage patterns and budgets now.

Given the specs and the current pricing, V4 Pro is arguably underpriced compared to similar models from other providers. But that's typical for DeepSeek—they've been aggressive on price since the beginning.

Bottom Line

DeepSeek V4 Pro is a serious upgrade for anyone doing agent work, coding assistance, or long-document analysis. The 1M context and 384K output are class-leading, and the compatibility with both OpenAI and Anthropic APIs removes friction. The concurrency cap is the only real limitation, but for most use cases it's a non-issue.

Is it time to switch? If you're already on DeepSeek's Flash model and hitting its limits, Pro is the obvious next step. If you're coming from another provider, the pricing and feature set make it worth a test run. Just keep an eye on those rates—the promised price hike might still be coming.