DeepSeek V4 lands quietly, with 75% promo price cut to follow
TL;DR:
- DeepSeek-V4-Pro shows a benchmark uplift over previous versions but, per Artificial Analysis data, ranks among leading open-weight models rather than clearly surpassing rivals like Kimi and Qwen.
- The Hangzhou startup followed up with a 75% promotional discount on V4-Pro through 5 May and cut input cache-hit prices across its API to one-tenth of original.
- For UK enterprise buyers, the more interesting question than benchmarks is whether V4’s Huawei-chip tuning and the ongoing US distillation accusations make Chinese open-weights a more or less safe procurement choice over the next quarter.
DeepSeek’s preview of its long-anticipated V4 model has drawn a noticeably muted reaction compared with the global tech-share selloff that followed V3 and R1 last year, according to Reuters. Benchmark data from Artificial Analysis shows V4-Pro improving on previous versions but slotting in among other leading open-weight models rather than running away from them.
That ordinary reception is itself the story. Analysts told Reuters the conditions that produced last year’s “black swan” — lofty US tech valuations, expectations of frontier-lab dominance, and surprise from a low-cost Chinese entrant — are no longer in place. “The expectation that new players will emerge is now baked into valuations,” Omdia’s Lian Jye Su said. Within China, multiple firms now release competitive models on a regular cadence, eroding DeepSeek’s relative lead.
A pricing follow-up the next morning
A few hours after the V4 launch, the company announced a 75% discount on V4-Pro until 5 May, alongside a cut in input cache-hit prices across the entire DeepSeek API to one-tenth of original, per a separate Reuters note. V4 ships in two variants — Pro for higher-capability work and Flash for lighter tasks — with DeepSeek positioning both as suited to AI agent workloads. The Pro version trails only Google’s Gemini-Pro-3.1 in world-knowledge benchmarks, the company says.
Looking forward
For UK enterprise readers, the V4 launch matters less for raw capability than for two adjacent questions. First, V4 is explicitly tuned for Huawei chip technology, accelerating the China-side compute-stack split that US export controls are designed to force. Second, the US State Department this weekend formalised a global cable warning about alleged Chinese “distillation” of US AI models, naming DeepSeek directly. Combined, those signals suggest the procurement posture for Chinese open-weight models in UK enterprise stacks is hardening — not because the models got worse, but because the political risk surface around them got wider. The right question for UK CIOs is no longer just “does the benchmark pass?” but “what happens to this dependency if US export rules tighten in the next three months?”