Per Artificial Analysis, DeepSeek V4 Flash scored 50 on its intelligence index,
ranking 3rd of 101 models and about twice the median among open-weight models.
Output speed is ~104 tokens/sec; input and output pricing are $0.14 and $0.28
per mln tokens. The report says competition is shifting from peak performance to
cost-per-completed-task. Low price, open weights and mln-token context position
DeepSeek for batch search, coding assistance and multi-agent calls, and will
pressure high-priced closed models to defend premiums via reliability, ecosystem
and enterprise services. Caveats: DeepSeek generated ~210 mln tokens during the
review—roughly double the peer median—suggesting verbose inference, and
benchmarks do not fully capture complex-task success rates; a low token price
may not translate into proportional cost savings for all real workloads.
