Skip to content
TheChatArchiveThe Chat Archive
o4-mini-2025-04-16· OpenAIHugging Face LMArena Human Preference 140K

what would be the cost

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with chatgpt-4o-latest-20250326’s answer.
  1. User

    what would be the cost

  2. Assistant

    Here’s a rough breakdown of the compute‐only cost if you build on top of DeepSeek-R1, using H100 spot/ondemand rates:

    Assumed H100 rate Spot – 1.00/hr∣On−demand–1.00/hr | On-demand – 2.00/hr

    PhaseH100-hrsCost @ $1/hrCost @ $2/hr
    Domain Adaptation / Fine-tune1 000–20 000$1 000–20 000$2 000–40 000
    Sim-to-Real Transfer10 000–100 000$10 000–100 000$20 000–200 000
    RL / Policy Optimization10 000–100 000$10 000–100 000$20 000–200 000
    Real-World Fine-tuning1 000–10 000$1 000–10 000$2 000–20 000
    Total22 000–230 000$22 000–230 000$44 000–460 000

    Key takeaways:

    • At 2/hr∗∗,expect∗∗2/hr**, expect **44K–$460K of GPU‐only costs.
    • Using spot instances (≈ $1/hr) or negotiated enterprise pricing can roughly halve this.

    Keep in mind this excludes:
    • Data storage, preprocessing, and networking fees
    • Engineering time (data collection, annotation, integration)
    • On-prem hardware depreciation (if you buy your own H100s)

    But purely for cloud GPU time, plan on tens to a few-hundred thousand dollars.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.