Week volume is Flash 0731 at 11.3T (+12% week-over-week, not share). The denser coding card is V4 Pro 0813. 0423 builds stay on this hub as lineage, not second cards.
OpenRouter This Week LLM Leaderboard tokens (prompt plus completion) through 18 Aug 2026: 11.3T on DeepSeek V4 Flash 0731 (+12% week-over-week, not share). Fetched 2026-08-19. Text request share for deepseek: 26.3% (418M requests) in the same window. That is request share, not tokens. OpenRouter This Week tokens are prompt plus completion. The percent next to a row is week-over-week change, not share of the week. Market share on this page is text request share by author, a different number.
Current Flash weights. 11.3T this week (+12% WoW). Continuum hosted id deepseek-v4-flash.
Current Pro weights. 946B this week, marked new. Not on the Continuum host list.
4.78T this week (−13% WoW). OpenRouter id deepseek/deepseek-v4-flash. Preview weights, not a second card.
2.57T this week (+1% WoW). OpenRouter id deepseek/deepseek-v4-pro. Preview weights, not a second card.
R1: 2026-08-19 OpenRouter row.
DeepSeek V3.2: 2026-08-19 OpenRouter row.
DeepSeek V3.2 Exp: 2026-08-19 OpenRouter row.
DeepSeek V3.1 Terminus: 2026-08-19 OpenRouter row.
DeepSeek V3.1: 2026-08-19 OpenRouter row.
R1 0528: 2026-08-19 OpenRouter row.
DeepSeek V3 0324: 2026-08-19 OpenRouter row.
R1 Distill Llama 70B: 2026-08-19 OpenRouter row.
DeepSeek V3: 2026-08-19 OpenRouter row.
Density proof for open weights. 0813 current at 946B this week. 0423 lineage.
Week-volume Flash. Hosted. Official Flash peak/off-peak, not Pro rates.
R1: 2026-08-19 OpenRouter row.
DeepSeek V3.2: 2026-08-19 OpenRouter row.
DeepSeek V3.2 Exp: 2026-08-19 OpenRouter row.
DeepSeek V3.1 Terminus: 2026-08-19 OpenRouter row.
DeepSeek V3.1: 2026-08-19 OpenRouter row.
R1 0528: 2026-08-19 OpenRouter row.
DeepSeek V3 0324: 2026-08-19 OpenRouter row.
R1 Distill Llama 70B: 2026-08-19 OpenRouter row.
DeepSeek V3: 2026-08-19 OpenRouter row.
This hub lists every published card. Start with DeepSeek V4 Flash.
Ranked view: Best coding models: the independent leaderboard puts every DeepSeek card with an independent coding row next to every other lab on one board, with the confidence intervals left visible.
The app is free with your own keys. Plus adds Continuum-hosted inference and the ids we actually list on each card.