DeepSeek API Price per Million Tokens vs Frontier Western Labs
| Model | Input ($/1M tokens) | Output ($/1M tokens) | Relative to frontier tier |
|---|---|---|---|
| DeepSeek V4-Flash | 0.14 | 0.28 | ~18x-36x cheaper input than GPT-5.5 / Opus 4.7 |
| DeepSeek V4-Pro (current, from 1 Jun 2026) | 0.435 | 0.87 | ~1/11 of GPT-5.5 input; was a launch promo, made permanent 1 Jun 2026; cache-hit input ~$0.0044/1M |
| DeepSeek V4-Pro (original launch price, pre-Jun 2026) | 1.74 | 3.48 | Original reference price at April 2026 launch; superseded by permanent cut |
| OpenAI GPT-5.5 (reference) | 5 | 30 | Frontier high-capability tier |
| Anthropic Claude Opus 4.7 (reference) | 5 | 25 | Frontier high-capability tier |
Updated at 2026-06
DeepSeek-V3 Disclosed Training Cost: Headline Claim vs Analyst Dispute
| Figure | Value | Source / status |
|---|---|---|
| Official V3 training run (GPU-hours) | ~2.788M H800 GPU-hours | DeepSeek-V3 technical report (disclosed) |
| Official V3 base-model cost (at $2/GPU-hr) | ~$5.576M | DeepSeek-disclosed; excludes prior research, ablations, failed runs |
| R1 reinforcement-learning phase | ~$0.294M | DeepSeek-disclosed (V3/R1 reporting) |
| Hardware acquisition (analyst est.) | ~$51M, RUMORED | SemiAnalysis estimate, not DeepSeek-confirmed |
| Total hardware outlay over company history (analyst est.) | Well above $500M, RUMORED | SemiAnalysis estimate, not DeepSeek-confirmed |
| Nvidia single-session market-cap loss (27 Jan 2025) | -$589B / -17% | Largest one-day market-cap loss for any US-listed company |
Updated at 2026-06
DeepSeek Funding and Valuation Trajectory
| Round (date, lead) | Amount Raised | Post-Money Valuation | Notable Investors |
|---|---|---|---|
| Founder / parent funding (2023-2025) | Undisclosed (off High-Flyer balance sheet) | n/a (wholly funded subsidiary) | High-Flyer Quantitative Investment Management (Liang Wenfeng) |
| Internal reference valuation (2026-04) | No round (mark only) | ~$10B (reported internal reference) | n/a |
| First external round (2026-06, RUMORED) | ~$7.4B (target ~50B yuan), RUMORED | $52B-$59B (reported, not closed), RUMORED | Tencent (~10B yuan), CATL (~5B yuan), NetEase (~3B yuan), JD.com (~3B yuan), Liang Wenfeng (~20B yuan) |
Updated at 2026-06
DeepSeek Model-Family Release Timeline and Parameter Scaling
| Date | Model | Total / active params, context | Significance |
|---|---|---|---|
| Nov 2023 | DeepSeek LLM (v1) | 7B and 67B dense (no MoE) | First open-weights release; base and chat variants |
| May 2024 | DeepSeek-V2 | 236B total / 21B active, 128K context | Introduced Multi-head Latent Attention (MLA); MoE efficiency thesis established |
| Dec 2024 | DeepSeek-V3 | 671B total / 37B active, trained on 14.8T tokens | The model behind the ~$5.6M / 2.78M H800 GPU-hour training-cost claim (figure heavily disputed) |
| Jan 2025 | DeepSeek-R1 | 671B total / 37B active | Open-weights reasoning model competitive with OpenAI o1; ~90.8 MMLU; triggered the Jan 27 2025 market shock |
| Dec 2025 | DeepSeek-V3.2 | MoE refresh | ~88.5 MMLU; bridge release ahead of V4 |
| Apr 24 2026 | DeepSeek-V4 (Pro + Flash) | V4-Pro 1.6T / 49B active; V4-Flash 284B / 13B active; 1M context | Near GPT-5.5 / Opus 4.7 coding quality at roughly one-sixth the API cost; optimized for Huawei Ascend |
Updated at 2026-04
Competitive-Programming Rating: DeepSeek V4-Pro vs Frontier Models (Codeforces)
Codeforces rating-Codeforces rating
Updated at 2026-05
DeepSeek V4-Pro vs Frontier Models on Coding Benchmarks
SWE-bench Verified (%)-LiveCodeBench
Updated at 2026-05
DeepSeek in AI Software
OpenAI
$PRIVATE
Largest AI lab by revenue. Approximately $13B annualized run-rate revenue (mid-2025) from a mix of ChatGPT consumer subscriptions, ChatGPT Enterprise / Team, the OpenAI API, and Microsoft Azure OpenAI revenue share. ChatGPT reaches roughly 800M weekly active users (late 2025). Builder of the GPT family (GPT-4o, GPT-4.5, GPT-5, o-series reasoning models) and Sora video. Microsoft is the dominant compute and distribution partner; SoftBank led the $40B March 2025 primary round at $300B.
