DeepInfra vs Groq: costs, performance and limits

DeepInfra costs $0.00054 per GPT-OSS 120B request example, 3.9× less than Groq’s $0.00210; Groq streams 8.4× faster at 473.4 vs 56.6 tokens/s.

Updated · 5 sources · Methodology

DimensionWinnerMarginEvidence
Estimated request costDeepInfra$0.00156$0.00054 / $0.002102345
Output speed (tokens/s)Groq8.4×56.6 / 473.4 tokens/s1
Endpoint Accuracy Index v1.0DeepInfra11 pp97% / 86%1

Performance

ProviderOutput speed (tokens/s)First-chunk median (s)Endpoint Accuracy Index v1.0
DeepInfra156.60.6297%
Groq1473.40.6386%

Prices and request cost

ProviderInput / 1M tokensOutput / 1M tokensEstimated request cost
DeepInfra23$0.037$0.17$0.00054
Groq45$0.15$0.60$0.00210
Cost example: 10,000 input + 1,000 billed output tokens
DeepInfra$0.00054
Groq$0.00210

Capabilities and serving conditions

DimensionDeepInfraGroq
API and model IDhttps://api.deepinfra.com/v1/openai
openai/gpt-oss-120b23		https://api.groq.com/openai/v1
openai/gpt-oss-120b45
Serving and capacityDefault account limit: 200 concurrent requests per model, counted as concurrency rather than per minute; increases are available on request.23Developer plan base limits for GPT-OSS 120B: 1K RPM, 500K RPD and 250K TPM; higher limits are available for select workloads. The free plan lists 30 RPM, 1K RPD, 8K TPM and 200K TPD.45

When each option fits

DeepInfra

  • Choose the standard endpoint when low recurring request cost matters more than draining each generated response quickly.23
  • Its 97% endpoint-accuracy reference is above Groq’s 86%, so the slower output is not accompanied by a lower accuracy reference here.1
  • DeepInfra has deprecated the Turbo listing for this model, so the standard endpoint is the one to select and price.23

Groq

  • Prefer Groq for output-heavy interactive responses: the matched AA reference is 473.4 tokens/s rather than DeepInfra’s 56.6.1
  • The fixed-token example is $0.00210 rather than $0.00054, so this speed choice increases the request budget.2345
  • Size production traffic against the Developer plan’s base limits of 1K RPM and 250K TPM for this model, not the free plan’s 30 RPM and 8K TPM.45

Methodology and sources

The saved AA provider snapshot compares GPT-OSS 120B at high reasoning effort with 10,000 input tokens. First-chunk latency excludes subsequent thinking and output time. Endpoint Accuracy Index is relative to a reference endpoint. Prices are official standard input/output rates, excluding cache discounts.

Costs are estimates at listed rates for uncached synchronous requests, not measured invoices; billed output includes reasoning. Tax, retries, discounts and negotiated capacity are excluded. Novita AI publishes this page using public sources, without a private workload benchmark.

  1. Artificial Analysis · GPT-OSS 120B Providers
  2. DeepInfra · GPT-OSS 120B
  3. DeepInfra · Rate Limits
  4. Groq · GPT-OSS 120B
  5. Groq · Rate Limits

Build with Novita Model APIs

Create an API key and connect your application.

Start BuildingExplore Models