Alibaba raises DeepSeek V4 Flash 0731 output price 230% to $1.32 per million tokens

Alibaba more than triples DeepSeek V4 Flash 0731 output pricing

Alibaba has raised the output price of DeepSeek V4 Flash 0731, a widely used model served through its API, from $0.40 to $1.32 per million tokens, a 230% increase. The figures come from the Stork AI newsletter dated October 20, 2026, which maintains an automated catalog that checks first-party list prices daily. According to Stork AI, the change landed overnight and without an announcement from the vendor, a pattern the outlet says is common when prices move upward.

Input pricing rose in the same update, from $0.20 to $0.44 per million tokens, more than doubling the cost of prompt tokens alongside the much larger jump on completions.

What the increase does to a bill

The arithmetic is punishing for output-heavy workloads. Ten million generated tokens cost $4.00 at the old rate and $13.20 at the new one. Applications built around agentic loops, where a single user action can trigger many model calls, multiply that difference across every step. Features that were comfortably profitable at the old price can flip into loss-makers where token spend is not tied directly to revenue.

Stork AI's assessment is that teams without caching layers, or those relying on verbose and unoptimized prompts, absorb the increase at full weight, while engineers who already manage token efficiency as a performance concern are far better insulated.

A wider wave of increases

The DeepSeek hike was not an isolated move. In the same tracking sweep, Stork AI recorded 17 distinct price movements across providers. Mistral raised its Ministral 3B model by 150% and its Ministral 8B by 50%, while Alibaba's own Qwen3.6 35B-A3B saw its output price climb 52%.

Only one provider moved sharply in the opposite direction: Zhipu AI cut the price of GLM-5V-Turbo by 85%, the sole major reduction in the batch.

A single-source reading, with caveats

These figures have not been matched against any vendor announcement, and Stork AI is the only tracker cited in the report. The story drew wider attention after reaching the Hacker News front page. For teams running the affected models at volume, the practical step is to verify current list prices directly with the provider before rebudgeting, since catalog data reflects list rates rather than negotiated enterprise discounts.

Why it matters

The unit economics of the AI application layer sit on top of API list prices that can change overnight and without notice. A 230% output increase on a widely used model shows how little pricing stability developers can assume, and how much vendor risk they take on when building on hosted models rather than infrastructure they control. The broader pattern recorded by Stork AI, with increases across most providers and one conspicuous cut, suggests inference is being repriced closer to what it actually costs to deliver, and that budgets premised on perpetually falling token prices need revisiting. For builders, caching, prompt efficiency and the ability to switch models quickly have shifted from optional optimizations to baseline engineering practice.