50 models7 providersprices scraped nightly — no estimates
deepseek · deepseek familyVerified May 27, 2026

DeepSeek V3

671B params · 131K context · deepseek · released 2024-12 · 4 providers

DeepSeek V3 is a 671-billion-parameter mixture-of-experts model from DeepSeek, released December 2024, routing each token through 8 of 256 experts for roughly 37B active parameters per forward pass. On general chat, math, and code benchmarks it punched well above its inference-cost weight at launch, drawing significant attention as a genuinely capable open-weights frontier model. Be aware that DeepSeek V3.2 shipped in May 2025 and dropped inference pricing roughly 30 percent, so V3 is now the legacy variant — still hosted on DeepInfra, Fireworks, and OpenRouter but no longer the cost-optimal choice within the V3 family. The 131K context window and DeepSeek's own license apply; verify commercial terms before use.

Cheapest right now
Hyperbolic$0.25/1M out
Go to Hyperbolic ↗Compare
sticky while scrolling · verified May 27, 2026
Cheapest now
$0.25
Hyperbolic · fp8 · $/1M out
90-day price Δ
—
cheapest-host trend
Fastest TTFT
480 ms
Fireworks AI · 150 tok/s
Providers
4
indexed providers
Ranks:#2Cheapest input#5Cheapest output#3Blended monthly cost#3Fastest TTFT#3Fastest throughput#3Context window#3MMLU score#3HumanEval score#3Provider count
✓ verified May 27, 2026
#Provider$/1M in$/1M outTTFT p50tok/sQuant
1HyperbolicCHEAPEST$0.25$0.25700 ms60fp8go ↗
2DeepInfra$0.20$0.85520 ms120fp8go ↗
3Fireworks AIFASTEST$0.25$1.00480 ms150fp8go ↗
4OpenRouter$0.27$1.10——unknowngo ↗
5DeepInfra$0.20$0.85520 ms120fp8go ↗
spread $0.25 → $1.10Cost for your workload →CSV ↓

Similar models

same class · by param size or family
ModelParamsFrom $/1M outFamily
DeepSeek V3.2671B—deepseekcompare →
Arctic Instruct480B—arcticcompare →
DeepSeek R1 Distill Llama 70B70B—deepseekcompare →
Yi 1.5 9B Chat9B—yiview →

Run it locally instead?

fp8~1476.2 GB VRAM
unknown~1476.2 GB VRAM
Which rigs run this →

Hosted at $0.25/1M on Hyperbolic. For most workloads, hosted is cheaper than local GPU power.

Questions developers ask

How much does it cost to run DeepSeek V3 for 100M tokens?▾

Running DeepSeek V3 with 100M input and 10M output tokens per month costs approximately $27.50 on Hyperbolic, the cheapest available provider as of the latest pricing data. Costs vary significantly depending on your input/output ratio and whether you use prompt caching.

What is the cheapest provider for DeepSeek V3?▾

Hyperbolic currently offers DeepSeek V3 at the lowest total cost for a standard workload. Prices change frequently — check the provider table above for the latest data.

What context window does DeepSeek V3 support?▾

DeepSeek V3 supports a context window of 131,072 tokens. Individual providers may cap this lower.

What's the cheapest way to run DeepSeek V3?▾

The cheapest way to run DeepSeek V3 is via Hyperbolic, starting at $0.25 per million input tokens. If your workload is prompt-heavy, enabling prompt caching can reduce costs further.

Is there a free tier for DeepSeek V3?▾

Free tiers vary by provider and change frequently. Check each provider's current pricing page for trial credits or free-tier limits. The prices shown on this page reflect paid API access.

How much does DeepSeek V3 cost for 1M tokens?▾

DeepSeek V3 input pricing starts at $0.25 per million tokens on Hyperbolic. Output tokens are typically priced 2–4× higher than input tokens depending on the provider.

Keep exploring

Cheapest on DeepInfra →Compare vs DeepSeek Coder V2 Instruct →Compare vs DeepSeek R1 →Compare vs DeepSeek R1 Distill Llama 70B →See full ranking: Cheapest input →See full ranking: Fastest TTFT →Used in the Customer support chatbot workload →Price history →
Methodology: prices scraped nightly from public pricing pages; snapshots append-only. Read methodology · Public API · CSV export · OG image per model.

Prices verified May 27, 2026 by scraping 4 providers. Methodology · Raw data