According to Stork AI’s daily LLM price catalog, which tracks first-party list prices across major AI labs, Alibaba raised the output price of DeepSeek V4 Flash 0731 from $0.40 to $1.32 per million tokens overnight—a 230% increase. Input costs more than doubled, rising from $0.20 to $0.44 per million tokens.
The price shock extends beyond DeepSeek. Mistral increased its Ministral 3B model by 150% and its 8B variant by 50%. Alibaba’s Qwen3.6 35B-A3B saw a 52% output price increase. Stork’s trackers recorded 17 distinct price movements across the field in a single night. The sole major exception was Zhipu AI, which cut GLM-5V-Turbo prices by 85%—moving counter to the broader trend.
For developers and companies relying on these models, the financial implications are immediate and severe. A 230% output token increase converts profitable features into loss leaders instantaneously. Any application running high-volume inference at current margin assumptions faces a dramatically rewritten economics.
According to the reporting, this represents a fundamental shift in market dynamics. The era of subsidized intelligence—where venture capital effectively paid users to consume API tokens—is closing. Labs have historically avoided issuing press releases for price increases, which is why these changes often catch builders by surprise.
The underlying driver appears to be infrastructure scarcity and rising operational costs. Concurrent reporting notes that SF Compute successfully executed a $245 million compute sale, underscoring sustained demand for guaranteed capacity. As labs and enterprises scale training and inference workloads, the scramble for computational resources has tightened, and those costs are now cascading to end users.
Builders face immediate pressure to optimize token efficiency. Strategies include implementing aggressive output caching, refining prompts to reduce verbosity, and reevaluating agentic loops that consume tokens without generating measurable revenue. Infrastructure engineers who treat token efficiency as a core performance metric now hold competitive advantage.
The price increases expose the fragility of businesses built entirely on subsidized API access. Companies that failed to develop alternative cost-reduction strategies or internal optimization layers are the immediate losers. The winners are those with in-house infrastructure or proven token-efficiency practices.
Key facts
- Alibaba raised DeepSeek V4 Flash 0731 output prices from $0.40 to $1.32 per million tokens (230% increase)
- DeepSeek V4 Flash input prices doubled from $0.20 to $0.44 per million tokens
- Mistral increased Ministral 3B prices by 150% and 8B prices by 50%
- Alibaba’s Qwen3.6 35B-A3B saw a 52% output price increase
- Zhipu AI cut GLM-5V-Turbo prices by 85%, the only major downward movement
- Stork AI tracked 17 distinct price movements across labs in a single night