DeepSeek V4
DeepSeek V4 is a large language model first observed on February 23, 2026, from the Chinese AI developer DeepSeek. The entity is also referenced under the aliases DeepSeek v2, DeepSeek-V4, and V4 model, indicating possible naming inconsistencies in early documentation. At launch, the company published commercial API pricing of $0.30 per million input tokens and $0.50 per million output tokens. As of late February 2026, DeepSeek has not released architecture details, parameter counts, training data specifications, or standardized benchmark results for independent verification, leaving its technical profile undefined against contemporaneous models. This release matters because it represents the latest move in an accelerating price war among foundation-model providers, with DeepSeek V4’s token costs positioned below many Western competitors at announcement. For enterprises and developers evaluating inference providers in early 2026, the model offers a low-cost option whose real-world performance remains unmeasured by third-party audits, making procurement decisions dependent on undisclosed capabilities.
Signal Radar
Five-axis snapshot of this entity's footprint
Mentions × Lab Attention
Weekly mentions (solid) and average article relevance (dotted)
Timeline
10- Research MilestoneJul 7, 2026
Research paper published introducing FlashMemory-DeepSeek-V4 with Lookahead Sparse Attention for ultra-long context
- paper:
- FlashMemory-DeepSeek-V4: Lightning Index Ultra-Long Context via Lookahead Sparse Attention
- Research MilestoneJun 9, 2026
DeepSeek V4 technical disclosure: 500K context with 90% less KV cache using FlashMemory
View source - Product LaunchMay 22, 2026
DeepSeek v4 API pricing permanently cut 75% to $0.43/M input tokens and $0.87/M output tokens
View source- amount:
- $0.43/M input, $0.87/M output
- round:
- permanent price cut
- Product LaunchMay 10, 2026
DeepSeek v4 launched, exposing performance gap in AMD ROCm vs NVIDIA CUDA
View source - Research MilestoneApr 13, 2026
Deployment of DeepSeek V4 on Huawei's Ascend AI chips discussed as part of China's semiconductor counterstrategy.
View source - Product LaunchApr 7, 2026
Began limited gray-scale testing with a new tiered interface (Fast, Expert, Vision modes).
View source - Product LaunchApr 3, 2026
Anticipated launch of DeepSeek V4 model powered by Huawei Ascend 950PR chips
View source - Product LaunchMar 1, 2026
Impending release of next-generation AI model, described as a 'whale' suggesting massive, powerful, disruptive potential.
View source - Research MilestoneMar 1, 2026
Previous model demonstrated competitive performance across multiple benchmarks with relatively modest computational requirements.
View source - Product LaunchJan 1, 2025
DeepSeek V4 released in early 2025
View source- parameters:
- 1.5 trillion
- architecture:
- Mixture-of-Experts
Relationships
14Developed
Uses
Competes With
Licensed
Frequently appears with
8Entities that show up in the same articles — shared coverage, not a stated relationship.
Recent Articles
2NVIDIA Blackwell Cuts DeepSeek V4 Token Costs 5x in One Month
+NVIDIA claims Blackwell inference stack cut DeepSeek V4 token costs 5x in one month, per a newly published report shared by @rohanpaul_ai.
100 relevanceMicrosoft Ditches Unlimited Copilot Tokens, Taps DeepSeek V4 for Cost Cuts
+Microsoft switched Copilot Cowork to usage-based pricing, adopting DeepSeek V4 to cut inference costs by ~40%. The move breaks Microsoft's exclusive r
95 relevance
Predictions
No predictions linked to this entity.
AI Discoveries
1- observationactiveJul 6, 2026
Lifecycle: DeepSeek V4
DeepSeek V4 is in 'declining' phase (0 mentions/3d, 1/14d, 19 total)
90% confidence
Sentiment History
| Week | Avg Sentiment | Mentions |
|---|---|---|
| 2026-W24 | 0.65 | 2 |
| 2026-W25 | 0.40 | 1 |
| 2026-W26 | 0.70 | 1 |
| 2026-W27 | 0.50 | 1 |