DeepSeek V4 nears full launch as Alibaba unveils Qwen 3.8

Two of China's most prominent AI labs are releasing flagship models within days of each other, sharpening a rivalry over pricing and performance that has reshaped the global AI landscape in 2026. DeepSeek is moving its V4 model family from preview to general availability this week, while Alibaba ↘ 4.74% has previewed its Qwen 3.8, a 2.4 trillion-parameter open-weight model it claims is "second only to Fable 5" among frontier systems.

DeepSeek V4 Reaches General Availability

DeepSeek first released V4 in preview form in late April, offering two tiers — V4 Pro, with 1.6 trillion total parameters and 49 billion active, and V4 Flash, a lighter 284-billion-parameter variant with 13 billion active. Both share a one-million-token context window and up to 384,000 tokens of output. The company announced in late June that a full general availability release would arrive in mid-July, accompanied by a new peak-and-valley pricing structure that charges double during high-traffic daytime hours.evolink+4

After a permanent price cut finalized in June, V4 Pro costs $0.435 per million input tokens and $0.87 per million output tokens, while V4 Flash runs at $0.14 and $0.28 respectively. DeepSeek says V4 Pro is roughly 12 times cheaper than OpenAI's GPT-5.5 on input tokens. Legacy model identifiers — "deepseek-chat" and "deepseek-reasoner" — will be retired on July 24. Early testers have reported strong improvements in agentic task execution, coding, and mathematical reasoning.technode+5

Alibaba Counters With Qwen 3.8

On July 19, Alibaba's Qwen team previewed Qwen3.8-Max, a multimodal model with 2.4 trillion parameters that processes text, images, video, and documents. Bloomberg reported that Alibaba shares rose following the announcement. The preview is available through Alibaba's Token Plan at 10 percent of standard pricing, with a nighttime discount reducing costs further. Open weights are promised "soon," though no date or license has been specified.the-decoder+4

Alibaba says Qwen 3.8 outperforms its predecessor Qwen 3.7-Max in coding, full-stack development, data analysis, and office workflows. Developer Shuai Bai described it as the team's first multimodal model exceeding one trillion parameters. However, independent benchmark results have not yet been published.marktechpost+2

A Crowded Field Gets Tighter

The near-simultaneous launches underscore an accelerating capability and pricing war among Chinese AI firms. Chinese-built open-source models have surged to more than 30 percent of traffic on the AI aggregation platform OpenRouter in some weeks this year, according to the Straits Times. The open-weight ecosystem — led by DeepSeek, Qwen, Meta's ↗ 1.02% Llama, and Mistral — has closed the gap to closed frontier models faster than most analysts expected.straitstimes+1

With DeepSeek betting on aggressive token pricing and Alibaba dangling the promise of freely downloadable weights for a model it says rivals frontier systems, the coming weeks will test whether either company can convert bold claims into sustained developer adoption.