13 Aug 2026
V4 Pro exits preview — runs OpenAI Codex natively
Flagship reasoning model goes GA at ~8x cheaper than comparable closed models, with native Codex support.
EN · Unite.AI ↗EN · AlphaSignal ↗
DeepSeek’s V4 line pairs flagship reasoning (V4 Pro) with a budget agent workhorse (V4.1 Flash) — open weights, 1M context, and off-peak pricing that undercuts most of the market.
DeepSeek is the Chinese lab that reset global price expectations with its open-weight MoE models. V4 Pro carries 1.6T total / 49B active parameters with a 1M context window; V4.1 Flash runs 284B / 13B for high-volume agent work.
| Model | What it is for | Vendor list price |
|---|---|---|
| DeepSeek V4 Proflagship · GA 13 Aug 2026 · 1M ctx | Reasoning and agentic MoE (1.6T/49B active) — 80.6% SWE-bench Verified, runs Codex natively | US$0.66 / US$1.98 per 1M off-peak (≈ RM2.69 / RM8.08) · peak US$1.32 / US$3.96 |
| DeepSeek V4.1 Flashbudget agent · GA 10 Sep 2026 · 1M ctx | 284B/13B active for high-volume agent tasks, with native vision | US$0.15 / US$0.60 per 1M off-peak (≈ RM0.61 / RM2.45) · peak US$0.30 / US$1.20 |
Last updated — Off-peak/peak billing since 17 Aug 2026 (peak = 2x off-peak, on input and output alike). Approx RM at 1 USD ≈ RM4.08 (spot, 19 Sep 2026). Cache-hit input US$0.003–0.044 /1M off-peak. On BD LLM Token Hub today. Figures are vendor public list prices in USD per 1M tokens, standard tier — no batch, cache, promo or peak-hour discount applied. Vendors change list prices without notice: confirm on the vendor’s own pricing page before budgeting.
Discontinued or fully superseded models, kept for historical reference. Active models — even previous-generation ones still on sale — always stay in the lineup above.
| Model | Era & what it was for | Historic list price |
|---|---|---|
| DeepSeek R1Jan 2025 · reasoning era | The open reasoning model that shook the market | US$0.55 / US$2.19 per 1M (≈ RM2.24 / RM8.94) — historic list |
| DeepSeek V3Dec 2024 | The original V3 general MoE | US$0.27 / US$1.10 per 1M (≈ RM1.10 / RM4.49) — historic list |
Recent announcements with outlet links (EN first, CN where available).
Flagship reasoning model goes GA at ~8x cheaper than comparable closed models, with native Codex support.
EN · Unite.AI ↗EN · AlphaSignal ↗DeepSeek moved to time-based pricing — off-peak is half of peak on both input and output, so agent schedules can be shaped around cost. Current per-model rates are in the table above.
EN · DeepSeek pricing ↗EN · AI Pricing Guru ↗The budget agent model is now DeepSeek-V4.1-Flash (model id deepseek-flash) with native vision, billed off-peak at US$0.15/M input and US$0.60/M output. The retired deepseek-v4-flash name routes to it at the Flash price.
One partner for every digital layer – explore what else we build, host and grow for Malaysian businesses.