inference/research
verifiedmedium · 0.72updated 2026-10-15

Hyperbolic AI token inference API

other

Fields

id
services/svc-hyperbolic
Provider
other
Endpoint
https://api.hyperbolic.xyz/v1
Input $/Mtok
0.12
Output $/Mtok
0.3
Price basis
USD per 1M TOKENS, representative model meta-llama/Llama-3.3-70B-Instruct at $0.12/$0.30, observed 2026-10-15 via LiteLLM's Hyperbolic provider table (hyperbolic.ai/pricing and /models both 404 on static fetch; api.hyperbolic.xyz/v1/models is auth-gated, so the aggregator transcription is the reachable public record). This is the joint-cheapest 70B-class input rate in data/services/ alongside DeepInfra Turbo's $0.10 floor and Novita's $0.135 - and HALF the per-token price Hyperbolic's own GPU-hour side implies, which is the point of a token API. Hyperbolic prices its reasoning and flagship entries flat same-in/out (DeepSeek-R1 $0.25/$0.25, Kimi-K2 and Qwen3-235B $2.00/$2.00); the $0.12/$0.30 headline is the standard-chat tier. provider='other' per the enum; the record name carries the provider.
Context window
131,072
Regions
us
notes
The other cheap-API outlier. $0.12 input on Llama 3.3 70B matches DeepInfra's floor-tier and undercuts Together ~8.7x on input; it lands here because Hyperbolic runs its own GPU fleet (H100/H200/B200 on-demand from $1.49/GPU-hr) rather than reselling someone else's. The catalog is small (~16 models on the LiteLLM transcription) and leans open-weight - Hermes-3-70B, Llama-3.1-405B, Qwen2.5-72B at the same $0.12/$0.30 tier, so the representative price generalises across the chat catalogue in a way OpenRouter's per-provider spread does not. No third-party measured TTFT/tok-s was reachable on 2026-10-15 (OpenRouter's provider table does not carry Hyperbolic), so measured_* stay null rather than borrow a marketing figure. Hermes-3-Llama-3.1-70B is served, but the repo model slug for the 70B Hermes is not in data/models/ (only the 8B distill is), so models_served names the closest repo-resolvable entries.

Notes

The other cheap-API outlier. $0.12 input on Llama 3.3 70B matches DeepInfra's floor-tier and undercuts Together ~8.7x on input; it lands here because Hyperbolic runs its own GPU fleet (H100/H200/B200 on-demand from $1.49/GPU-hr) rather than reselling someone else's. The catalog is small (~16 models on the LiteLLM transcription) and leans open-weight - Hermes-3-70B, Llama-3.1-405B, Qwen2.5-72B at the same $0.12/$0.30 tier, so the representative price generalises across the chat catalogue in a way OpenRouter's per-provider spread does not. No third-party measured TTFT/tok-s was reachable on 2026-10-15 (OpenRouter's provider table does not carry Hyperbolic), so measured_* stay null rather than borrow a marketing figure. Hermes-3-Llama-3.1-70B is served, but the repo model slug for the 70B Hermes is not in data/models/ (only the 8B distill is), so models_served names the closest repo-resolvable entries.

Cross-references

Models (6)