Open weights model released October 2026.
Benchmark summary (Artificial Analysis)
- Intelligence Index: #2 / 121 models — score 45 on Artificial Analysis Intelligence Index v4.3.2
- Speed: #6 / 79 — 230.4 output tokens per second
- Pricing: 1.20 per 1M output tokens, 98% cache discount; ~$0.26 cost per Intelligence Index task
- Verbosity: Very verbose (210M output tokens on Intelligence Index vs median 140M)
Comparison summary: DeepSeek V4.1 Flash 1017 (Max) is amongst the leading models in intelligence and reasonably priced among open-weight models of similar size; notably fast but very verbose. Supports text and image input, text output, 1M token context window.
Technical specifications
| Field | Value |
|---|---|
| Reasoning | Yes (reasoning variant; non-reasoning may exist) |
| Input modality | Text and image |
| Output modality | Text |
| Context window | 1M tokens (~1500 A4 pages) |
| Total parameters | 552B |
| Active parameters | 16B per token |
| License | MIT |
| Model weights | Hugging Face |
Intelligence Index v4.3.2 incorporates: AA-Briefcase v1.1, GDPval-AA v2.1, AutomationBench-AA, Terminal-Bench 4.0, SciCode, Humanity’s Last Exam, GDP.pdf, CritPt, AA-Omniscience, AA-LCR v1.1.
Note: Page indicates no publicly available API for this model listing at time of scrape.