Skip to content

Provider Reference

Noxaudit supports three AI providers. You can rotate between them to get different perspectives on your codebase.

Provider Comparison

Provider Model Input $/M Output $/M Context Batch Discount Notes
Anthropic claude-opus-4-6 $5.00 $25.00 200K 50% Most findings, best depth. Tiered pricing above 200K
Anthropic claude-sonnet-4-6 $3.00 $15.00 200K 50% Best precision (6/6 consensus). Tiered pricing above 200K
Anthropic claude-haiku-4-5 $1.00 $5.00 200K 50% Budget Anthropic option
Google gemini-2.5-pro $1.25 $10.00 1M 50% Large context, tiered pricing above 200K
Google gemini-3-flash-preview $0.50 $3.00 1M 50% Latest Gemini flash (preview)
Google gemini-2.5-flash $0.30 $2.50 1M 50% Pre-pass triage model
OpenAI gpt-5.4 $2.50 $15.00 1M 50% Best mid-tier value (84 findings in benchmark)
OpenAI o4-mini $1.10 $4.40 200K 50% Reasoning model — poor cost efficiency for auditing
OpenAI gpt-5-mini $0.25 $2.00 400K 50% Best daily value ($0.03/run, 5/6 consensus)
OpenAI gpt-5-nano $0.05 $0.40 400K 50% Cheapest — too shallow for most use cases

Benchmark-Informed Recommendations

We benchmarked all 10 models against two repos (python-dotenv + noxaudit itself) and validated quality using cross-model consensus. See the full Benchmark Results for details.

Key findings:

  • gpt-5-mini ($0.03/run) hits 5/6 consensus issues with minimal noise — best daily value
  • gpt-5.4 ($0.26/run) found 84 total findings and beats Sonnet quality at 68% the cost
  • claude-opus-4-6 ($0.65/run) found the most findings overall — best for maximum depth
  • claude-sonnet-4-6 ($0.38/run) hit 6/6 consensus with the lowest noise — best precision
  • Gemini models had lower consensus scores (2-3/6) but offer 1M context for large codebases
  • o3 was dropped entirely (0 findings on the canary repo)

Tiered Pricing

Anthropic and Gemini 2.5 Pro use tiered pricing — input tokens above 200K are billed at 2x the standard rate, and output tokens also switch to the higher tier:

Model Standard Input High-Tier Input (>200K) Standard Output High-Tier Output
claude-opus-4-6 $5.00/M $10.00/M $25.00/M $37.50/M
claude-sonnet-4-6 $3.00/M $6.00/M $15.00/M $22.50/M
gemini-2.5-pro $1.25/M $2.50/M $10.00/M $15.00/M

Tip

Use noxaudit estimate to check if your codebase exceeds the 200K tier threshold. If it does, consider using the pre-pass to reduce token count, or switch to a Gemini model with a 1M context window.

Cache Token Pricing (Anthropic)

Anthropic supports prompt caching, which reduces costs on repeated runs:

Model Cache Read $/M Cache Write $/M
claude-opus-4-6 $0.50 $6.25
claude-sonnet-4-6 $0.30 $3.75
claude-haiku-4-5 $0.10 $1.25

Cache read tokens cost 10% of standard input price. Cache write tokens cost 125% of standard input price.

Setup per Provider

Anthropic

pip install noxaudit  # included by default
export ANTHROPIC_API_KEY=sk-ant-...

repos:
  - name: my-app
    provider_rotation: [anthropic]
model: claude-sonnet-4-6

Anthropic uses the batch API by default, which provides a 50% discount. Use noxaudit submit / noxaudit retrieve for the batch workflow.

OpenAI

pip install 'noxaudit[openai]'
export OPENAI_API_KEY=sk-...

repos:
  - name: my-app
    provider_rotation: [openai]
model: gpt-5-mini

Google Gemini

pip install 'noxaudit[google]'
export GOOGLE_API_KEY=...

repos:
  - name: my-app
    provider_rotation: [gemini]
model: gemini-2.5-flash

Gemini's 1M context window makes it ideal for large codebases where Anthropic would hit tiered pricing.

Multi-Provider Rotation

Rotate between providers to get different perspectives:

repos:
  - name: my-app
    provider_rotation: [anthropic, openai, gemini]

Noxaudit cycles through the list on each run. Different models catch different things — our benchmark showed only 6 issues had consensus across models, confirming that rotation improves coverage.

Based on our benchmark results, a tiered approach works best:

Use Case Provider Model Cost/Run Why
Daily audits OpenAI gpt-5-mini ~$0.03 5/6 consensus, minimal noise
Deep dives OpenAI gpt-5.4 ~$0.26 84 findings, beats Sonnet at 68% cost
Premium analysis Anthropic claude-opus-4-6 ~$0.65 Most findings, best for max depth