DeepSeek V4.1 Flash
ModelAPIOpen weights (MIT)
What it is
DeepSeek V4.1 Flash is an MIT-licensed open-weights multimodal MoE model with a 1M-token context, served on the DeepSeek API as deepseek-flash.
Best for
#2 of 3Model for Build a deep-research agent
Total 7.4 / 10
- Task quality (40%)6 / 1085.8% on BrowseComp per an aggregator only; unverified at source.
Evidence:0118 (source: The Aggregate)
- Cost per typical job run (20%)9 / 10$0.30/$1.20 per MTok peak (off-peak half); $0.27 per AA index task.
Evidence:0122 (source: Artificial Analysis)0121 (source: DeepSeek)
- Latency / speed (5%)8 / 10175.9 output tokens/s.
Evidence:0122 (source: Artificial Analysis)
- Tool-calling & structured-output reliability (15%)8 / 1076.8% on τ²-Airline.
Evidence:0083 (source: OpenRouter)
- Context & modality support (10%)8 / 101M context; image + text input.
Evidence:0119 (source: Hugging Face)
- Openness / deployability (10%)8 / 10MIT open weights; self-hosting a 552B MoE needs serious hardware.
Evidence:0119 (source: Hugging Face)
Strengths
- MIT open weights.0119 (source: Hugging Face)
- $0.27 per Artificial Analysis index task; 175.9 tokens/s.0122 (source: Artificial Analysis)
Limits
- BrowseComp score only seen on an aggregator, not at source.0118 (source: The Aggregate)
- 552B-parameter MoE: self-hosting needs large GPU servers.0119 (source: Hugging Face)
Pricing
Paid
Peak $0.30 in / $1.20 out per MTok; off-peak 50% off.0121 (source: DeepSeek)0120 (source: DeepSeek)
As of . Check the vendor's page before buying.
License
Open weights (MIT).0119 (source: Hugging Face)
Checked .
Alternatives in the same layer
Disclosures
- None