Event Summary
On April 24, 2026, DeepSeek released V4, its most powerful open-weight model family — with V4-Pro at 1.6 trillion total parameters (49B active) and V4-Flash at 284B (13B active), both under the MIT license with a 1-million token context window. Priced at $0.87 per million output tokens, V4 undercut GPT-5.5, Claude Opus 4.7, and Gemini 3.1 Pro by 5-10x. The release demonstrated that the open-weight ecosystem had permanently reshaped AI economics.
Context & Narrative
DeepSeek V4 was the continuation of a trajectory that began with DeepSeek-R1 in January 2025. Where R1 had shocked the industry by proving open-source could match closed-source reasoning, V4 extended that proof to scale: a 1.6 trillion parameter open-weight model that was competitive with the best closed-source models while being dramatically cheaper. The architecture continued DeepSeek's Mixture-of-Experts approach, with only 49B parameters active per token — making inference efficient despite the massive total parameter count. Key benchmarks showed V4-Pro competitive with GPT-5.5 and Claude Opus 4.7 across reasoning, coding, and mathematics. V4-Flash offered a smaller, faster option for high-volume deployments. The pricing — $0.87/M output tokens — was intentionally aggressive, designed to capture market share from US competitors. The release had three major implications. First, it validated DeepSeek's long-term strategy: the Chinese lab had gone from 'interesting open-source experiment' to 'permanent frontier competitor' in 16 months. Second, it continued the commoditization of frontier AI: if a 1.6T open-weight model cost $0.87/M, the era of $15-75/M pricing for closed models was unsustainable. Third, it intensified the US-China AI competition: DeepSeek's ability to produce frontier-competitive models despite US chip restrictions demonstrated the limits of export controls. The open-weight community embraced V4 immediately, with fine-tuned variants appearing on Hugging Face within hours of release.
Key Findings
-
Fact Grade C
DeepSeek released V4 on April 24, 2026 — an open-weight (MIT) model family with V4-Pro at 1.6T parameters and V4-Flash at 284B, priced at $0.87/M output tokens.
Sources [1]
Impact Assessment
-
Economic Disruption +2 · Medium-term
Priced at $0.87/M output tokens — 5-10x cheaper than GPT-5.5, Claude, and Gemini. Accelerated the commoditization of frontier AI. Put sustained downward pressure on AI model pricing across the industry.
Affected Groups: AI developers, startups, enterprises, AI companies
-
Access Democratization +1 · Medium-term
MIT-licensed 1.6T open-weight model made frontier-level AI accessible to anyone who could host it. Continued the DeepSeek lineage of making advanced AI available at dramatically reduced costs.
Affected Groups: developers, researchers, global south
Consensus & Sources
-
1
DeepSeek V4 is an open-weight mixture-of-experts model family released April 24, 2026 under the MIT license.Reference Evidence Citation logged Live source
-
2
DeepSeek-V4-Pro on Hugging Face.Reference Evidence Citation logged Live source
-
3
DeepSeek V4 Preview Release announcement.Reference Evidence Citation logged Live source