DeepInfra
Based on public pricing signals, xpay verified that DeepInfra exposes per-MTok rates for dozens of model families, per-image formulas for FLUX, and per-GPU-hour custom LLM rates on a public anonymous page. Self-serve usage tiers auto-promote on spend. xpay could not locate /.well-known/ai-pricing.json or anonymous agent purchase paths.
Simple Pricing, Deep Infrastructure
DeepSeek V3.1 $0.21/$0.79 per MTok; Llama 3.1 8B $0.02/$0.05; H100 custom-LLM $1.79/GPU-hr
$0
Self-serve
Self-serve
Self-serve
Self-serve
Contact sales
How DeepInfra scores on the 7 agent-ready dimensions
Public pricing
15 / 15
Usage-based / metered
22 / 25
Self-serve checkout
13 / 15
Public API
15 / 15
Low / no minimum
5 / 10
Unauth automated payment
6 / 10
Bonus (machine-readable pricing)On top of /100 base
0 / 5
Total
76 / 100
Six-step check: can an agent actually buy from DeepInfra?
Discover price
https://deepinfra.com/pricing — model-by-model tablesSelect a plan
Five Usage Tiers programmatically distinctPay per task
Per-token billing on every LLM and per-image formula on FluxAvoid a sales call
All self-serve; only DGX clusters gatedAPI docs without auth
https://docs.deepinfra.com publicEstimate cost upfront
Explicit per-MTok plus context length per modelPros and cons for AI agents
Observational summary written by xpay from the signals captured on 2026-05-03. Not a review of the product — only of its current pricing posture for agent buyers.- Pricing is publicly visible on an indexable page — agents can read tiers without scraping past auth.
- Per-unit billing is published, so an agent can budget for a single task before committing.
- API documentation is reachable without a login — discovery and integration can happen in one session.
- Per-unit rate is concrete enough that an agent can model expected spend before issuing a request.
- A free tier exists, lowering the bar for an agent builder to prototype before committing budget.
- Some tiers are sales-led; the highest-capacity surfaces are not self-serve.
- PPU requires account creation and an issued API key; a fully unauthenticated agent purchase is not yet supported.
- No /.well-known/ai-pricing.json or equivalent machine-readable pricing manifest is published — agents must rely on HTML scraping.
How DeepInfra could lift its score
DeepInfra is already Verified at 88. Remaining 12 points: publish /.well-known/ai-pricing.json mirroring the per-model rates, and add anonymous one-shot agent purchase via x402.
| pricing_visible | true |
| headline_phrasing | Simple Pricing, Deep Infrastructure |
| tier_count | 5 |
| lowest_paid_entry_usd | 0 |
| free_tier | true |
| free_tier_terms | Tier 1 starts with $20 invoicing threshold, PAYG |
| per_unit_price | DeepSeek V3.1 $0.21/$0.79 per MTok; Llama 3.1 8B $0.02/$0.05; H100 custom-LLM $1.79/GPU-hr |
| annual_required | false |
| self_serve_paid_tiers | 5 |
| sales_only_tiers | 1 |
| public_api_docs_url | https://docs.deepinfra.com |
| api_docs_auth_walled | false |
| ai_pricing_json_present | false |
| agents_txt_present | false |
| anonymous_purchase_path | false |
DeepInfra
76
/ 100 (rubric v1.1)AI Inference
Simple Pricing, Deep Infrastructure
5
Yes
Tier 1 starts with $20 invoicing threshold, PAYG
Free
DeepSeek V3.1 $0.21/$0.79 per MTok; Llama 3.1 8B $0.02/$0.05; H100 custom-LLM $1.79/GPU-hr
5 / 1
Public
No
Not published
2026-05-03