NewsBite analysis · AI, Cloud
ProPakistani
DeepSeek V4-Flash Beats GLM 5.2, Rivals Opus 4.8 While Costing Just $0.28
DeepSeek has launched its V4-Flash model, an open-source lightweight AI solution that surpasses notable models like GLM 5.2, while remaining cost-effective at just $0.28 per million output tokens. It also demonstrates competitive performance against Claude Opus 4.8 in various benchmarks. The model employs a sophisticated mixture-of-experts architecture with 284 billion parameters, allowing extensive context capabilities and effective reasoning modes tailored for different workloads. Its release under the MIT license facilitates broader adoption for commercial and on-premise usage. Developers are encouraged to evaluate its performance across their specific applications, considering that some results are from vendor reports.
Act nowConfidence: HighShort term horizon
Published 3 Aug 2026, 11:55 UTC4 min read
Visit original article ↗Lead image — ProPakistani
Summary
TL;DR
- V4-Flash beats GLM 5.2 in all nine benchmark tests.
- Costs $0.28 per million output tokens, significantly lower than competitors.
- Supports a one-million-token context window.
- Open-sourced under MIT, allowing for flexible commercial use.
- Utilizes a mixture-of-experts architecture for enhanced performance.
Interesting facts
- V4-Flash includes a DSpark speculative decoding feature for faster token generation.
- It has an advanced configuration with 13 billion activated per token.
- DeepSeek trained V4-Flash on over 32 trillion tokens.