DeepSeek V4-Flash Enters Public Beta, and OpenAI Slashes GPT-5.6 Prices

+

On one side, DeepSeek's V4-Flash, with stronger Agent capabilities, has stormed into public beta; on the other, OpenAI has cut its flagship model to 20% of its price, unprecedented in its history. This week, the competition over AI inference and Agents has fully entered white-hot territory.

1. DeepSeek V4-Flash: A Full Leap in Agent Capability

DeepSeek has officially announced that the official API for DeepSeek-V4-Flash has entered public beta. This is not just another version iteration; it's more like a clear declaration of stance: in the Agent race, DeepSeek is going head-on.

Two things about this upgrade are worth noting:

1. Agent capability is greatly enhanced, surpassing V4-Pro-Preview across the benchmarks

V4-Flash is positioned as "fast and strong". Official figures show its scores on several Agent-related benchmarks already far exceed the previous-generation Pro preview. For actual users, this means that in high-frequency Agent scenarios like tool calling, multi-step reasoning, and code execution, V4-Flash is not just "faster" but "stronger".

2. Native support for the Responses API format, fully compatible with Codex

This is the most critical piece of information. The Responses API is a next-generation interface spec proposed by OpenAI. DeepSeek now supports it natively, and its official docs directly provide a tutorial for connecting to Codex, which amounts to saying: developers can painlessly drop DeepSeek into Agent workflows originally built for OpenAI.

Behind this is a clever move: at the key moment when the Agent toolchain ecosystem (especially coding Agents like Codex) is taking shape, trade "compatibility" for "entry". The cost to developers is zero, while the ecosystem payoff is direct.

2. OpenAI: Not a Price Cut, but a Clearing of the Field

At the same time, OpenAI wasn't idle either, and hit harder:

• GPT-5.6 Luna price cut by 80%, bringing the flagship-tier model straight down to 20% of its price;

• GPT-5.6 Terra cut by 20%;

• GPT-5.6 Sol gets a faster option in the API;

• More importantly, the price cuts on Luna and Terra are reflected directly in usage billing for Codex and ChatGPT Work, so the more users use, the more they save.

OpenAI's own words were "We are committed to pushing the model frontier across cost efficiency, capability, and speed." That is cost efficiency, capability, and speed, advancing on all three fronts.

Put the two things side by side and it gets interesting.

3. Two News Items, One Battlefield

On the surface it's "one release, one price cut", but both actually point to the same battlefield: the contest over model supply in the Agent era.

• DeepSeek's strategy is "entry through compatibility": with lower prices plus native Responses API support, it embeds itself into the ecosystem standard defined by OpenAI, making switching almost costless for developers.

• OpenAI's strategy is "clearing the field with price": it proactively cuts its flagship model by 80%, directly squeezing the price advantage of latecomers, while extending the price-cut dividend to its own products such as Codex and ChatGPT Work to strengthen ecosystem stickiness.

One is grabbing the entrance at the "compatibility layer", the other is blocking the exits at the "price layer". Neither mentions the other, yet every move lands on the other.

4. What This Means for Developers and Teams

1. The cost of switching models has never been lower. Once interfaces are compatible and formats are standardized, a team may only need to change one line of config to switch between DeepSeek and OpenAI, turning "multi-model fallback" from a slogan into an engineering reality.

2. The price war is far from over, and the fire has spread to Agent scenarios. When inference costs drop by orders of magnitude, the product shape of AI applications will be redefined: features that never added up financially in the past are now feasible.

3. Coding Agents are becoming a must-win ground. Codex is OpenAI's home turf, and DeepSeek has adapted to Codex directly; OpenAI in turn ties Codex users in with billing cuts. Whoever occupies developers' daily workflow gets the ticket to the next round of the ecosystem.

4. Meanwhile, the GLM plans have changed their metering mechanism and also reopened subscriptions, so there's no more need to scramble for them as before. The bad news is that plan prices have risen sharply, and judging from the DeepSeek public beta data, they now exceed GLM 5.2, so I'd guess everyone has new ideas and new choices.

+

The short-term winners of this round of clashes are developers: a stronger model, lower prices, and a smoother switching experience, all three in hand at once.

In the long run, the "model supply" for Agents is becoming infrastructure like water, electricity, and coal, and a price war in infrastructure often signals the real start of an industry's maturity.

Sources:

• Official DeepSeek tweet: https://x.com/deepseek_ai/status/2083084415157022911

• Official OpenAI tweet: https://x.com/OpenAI/status/2082878156483219672

• DeepSeek Codex integration docs: https://api-docs.deepseek.com/quick_start/agent_integrations/codex

+