DeepSeek Plans Price Hike as OpenAI Cuts GPT-5.6 Luna Costs and Goes Free for Millions

OpenAI’s Free Push Meets DeepSeek’s Price Hike

OpenAI announced on Aug. 6 that its GPT-5.6 Luna model will become free and unlimited for all Free and Go users, directly affecting roughly one billion accounts worldwide. The move follows an 80 percent price cut for the same model on July 30, when output prices fell from $6 to $1.20 per million tokens and input prices dropped from $1 to 20 cents. On the same day DeepSeek launched the official API for its V4 Flash model, and this week it warned customers that API prices would face a “significant” increase in the near future.

The announcements highlight the intensifying tug-of-war between proprietary Western AI firms and lower-cost, open-source Chinese competitors. Third-party aggregation data from OpenRouter shows that this week calls to DeepSeek V4 Flash reached 11.31 trillion tokens, nearly three times the 3.54 trillion tokens for GPT-5.6 Luna. Although the figures exclude direct developer calls to OpenAI, the gap suggests that previous price reductions haven’t reversed the flow of developers migrating to Chinese models.

According to a Huatai Securities report, DeepSeek V4 Flash costs about 6 US cents per million tokens at the package level — 65 percent cheaper than GPT-5.6 Luna — and the per-task cost is roughly 57 percent lower. With the performance gap narrowing, many developers have voted with their feet. OpenAI’s latest counter is the free tier, which it says reduces factual errors by 62 percent compared to its default free model, GPT-5.5 Instant, while the premium GPT-5.6 Sol remains the main revenue driver for paid subscribers.

On the other side, DeepSeek’s breakneck call volume growth has become a double-edged sword. The surge in AI‑agent tasks, which can require 100 times the computing power of a simple chat, strains infrastructure and erodes service stability. That has pushed several Chinese AI firms to recalibrate. Zhipu AI raised API prices 83 percent in the first quarter and still saw call volume jump 400 percent, and Alibaba, Tencent and Baidu all hiked AI‑related prices earlier this year. DeepSeek’s own pivot is part of a strategic shift from buying market share with low prices to charging based on the value delivered.

Behind the Price Reversal: Why DeepSeek Is Changing Course After Dominating Call Volumes

OpenAI’s Free Gambit: Hooking Developers and Consumers

Making the top‑of‑the‑line model free for a billion users is a defensive move against the gravitational pull of cheap open‑source alternatives. Previous price cuts failed to stem developer migration, so OpenAI is now betting that eliminating the cost barrier entirely — while raising factual accuracy — will create stickiness. In the short term this could commoditize the free tier further, but the company’s strategy clearly hinges on converting a fraction of those users to the paid GPT‑5.6 Sol for advanced reasoning and agentic tasks. For developers, the free tier removes a cost reason to switch, but lock‑in will rely on ecosystem integration and consistent performance.

DeepSeek’s Volume Trap: Why More Calls Can Mean More Losses

DeepSeek’s V4 Flash quickly topped the global call volume charts, but the economics of high‑volume, low‑price API services are punishing. Each AI‑agent task can consume 100 times the compute of a regular dialog, and peak‑time demand degrades service stability — a direct consequence of the low‑cost model. By raising prices, DeepSeek aims to break the cycle in which every additional call deepens the loss. The Zhipu AI precedent shows that a steep price rise doesn’t necessarily crater demand if the product is seen as high‑value; in fact, call volumes can continue to climb if reliability improves. DeepSeek is betting that its developer base will pay more for better service, but the risk is that price‑sensitive users move to OpenAI’s free tier or to other low‑cost Chinese alternatives that are still cheaper.

The Industrial Pivot: From Subsidized Growth to Profitable AI

The collective price increases by Alibaba, Tencent, Baidu and now DeepSeek signal that the phase of aggressive price wars in Chinese AI is winding down. The initial low‑cost strategies were designed to build developer ecosystems and data flywheels, but as AI workloads become more complex and compute costs multiply, the old economics are unsustainable. This shift mirrors the maturation of cloud computing, where loss‑leader pricing eventually gave way to margin discipline. For the broader industry, the era of ultra‑cheap API tokens may be ending, forcing developers to architect more cost‑efficient applications or consider self‑hosted open‑source models.

What Developers and Enterprises Need to Watch as AI API Prices Shift

• If your application relies on DeepSeek’s current API pricing, model a budget impact using a conservative increase and consider whether promised service reliability improvements justify the cost. DeepSeek has explicitly warned of a significant near‑term hike.
• Developers on OpenAI’s Free or Go tiers now have unlimited access to GPT‑5.6 Luna for text generation — a top‑tier model at no cost — reducing the economic incentive to use cheaper third‑party APIs for those workloads.
• Given that Zhipu AI, Alibaba, Tencent and Baidu have already raised prices, enterprises should plan for the closing window of ultra‑low‑cost Chinese API access and evaluate optimization or self‑hosting strategies.
• Track DeepSeek’s service stability metrics during peak hours; if the price rise also eases the compute bottleneck, high‑volume users may stay. If stability doesn’t improve, the cost increase could accelerate defections to reliable alternatives.