Tensorust AI Cost Audit

Sample: AI cost audit of an example usage file

Generated 2026-10-25 15:40 UTC from OpenAI usage export (150 rows). Prices: Tensorust price table, snapshot 2026-10-23, every price linked to the provider's pricing page. Your file covers 30 days (2026-09-24 to 2026-10-23); every volume is scaled to a 30.4375-day month (x1.015).

Savings are not additive: each one is measured against what you pay today. Every number is list price times your token volumes; your invoice can differ by negotiated discounts, credits, taxes, data-residency or regional uplifts and tool fees (web search, code execution), which a token export does not show.

1 model is not in our price table and is left out of every total: text-embedding-3-small.

Recommended actions

1. GPT-4o is marked "deprecated" by OpenAIPlan the move before it stops answering. The cheapest comparable current model for this volume is GPT-6 Luna at $96.94/month. Check the provider's deprecation page for the exact date.
2. GPT-4o mini is marked "deprecated" by OpenAIPlan the move before it stops answering. The cheapest comparable current model for this volume is GPT-6 Luna at $217/month. Check the provider's deprecation page for the exact date.
3. Test GPT-6 Luna for your GPT-4o traffic: $2,316 a month lower$2,413 now, $96.94 on GPT-6 Luna for the same tokens: $2,316 a month lower. Price difference only; test quality and latency before switching.
4. Test GPT-6 Luna for your GPT-5.2 traffic: $1,243 a month lower$1,302 now, $59.01 on GPT-6 Luna for the same tokens: $1,243 a month lower. Price difference only; test quality and latency before switching.
5. Send it through the batch API: GPT-4o: $1,206 a month lowerSame model, results within 24 hours instead of seconds. Only for work that can wait: evals, backfills, nightly jobs, extraction. $2,413 now, $1,206 in this scenario.
6. Send it through the batch API: GPT-5.2: $651 a month lowerSame model, results within 24 hours instead of seconds. Only for work that can wait: evals, backfills, nightly jobs, extraction. $1,302 now, $651 in this scenario.
7. Use the flex tier: GPT-5.2: $651 a month lowerSame model at the provider's flex tier: slower responses and occasional resource-unavailable errors, lower price. $1,302 now, $651 in this scenario.
8. Test GPT-6 Luna for your GPT-5.4 mini traffic: $438 a month lower$499 now, $60.88 on GPT-6 Luna for the same tokens: $438 a month lower. Price difference only; test quality and latency before switching.
9. Send it through the batch API: GPT-4o mini: $152 a month lowerSame model, results within 24 hours instead of seconds. Only for work that can wait: evals, backfills, nightly jobs, extraction. $305 now, $152 in this scenario.
10. Cache your repeated prompt prefix: GPT-4o: $128 a month lowerToday 35% of your input is billed as cache reads. Scenario: 50% of input served from cache (a fixed system prompt, tool definitions or shared documents put first). Your real hit rate depends on how much of each prompt repeats. $2,413 now, $2,285 in this scenario.
11. Cache your repeated prompt prefix: GPT-5.4 mini: $111 a month lowerToday 0% of your input is billed as cache reads. Scenario: 50% of input served from cache (a fixed system prompt, tool definitions or shared documents put first). Your real hit rate depends on how much of each prompt repeats. $499 now, $388 in this scenario.
12. Test GPT-6 Luna for your GPT-4o mini traffic: $87.57 a month lower$305 now, $217 on GPT-6 Luna for the same tokens: $87.57 a month lower. Price difference only; test quality and latency before switching.
13. Cache your repeated prompt prefix: GPT-4o mini: $39.78 a month lowerToday 10% of your input is billed as cache reads. Scenario: 50% of input served from cache (a fixed system prompt, tool definitions or shared documents put first). Your real hit rate depends on how much of each prompt repeats. $305 now, $265 in this scenario.

Where your money goes

Model by model

GPT-4o OpenAI standard deprecated

Your rows: "gpt-4o-2024-08-06" (matched without the date suffix). Price source: developers.openai.com/api/docs/pricing, checked 2026-10-23.

Rates applied per 1M tokens: input $2.5, cache read $1.25, cache write $2.5, output $10.

Cheapest comparable models for this exact volume Price difference only; test quality and latency before switching.

Same provider (OpenAI), comparable on paper: GPT-6 Luna $96.94 · GPT-5.6 Luna $214 · GPT-5.4 mini $802 · GPT-6.1 Sol $1,915 · GPT-6 Sol $1,939

Price history for this model: 2026-10-23: GPT-4o added to the table: $2.5 input / $10 output per 1M tokens.

GPT-5.2 OpenAI standard

Your rows: "gpt-5-2". Price source: developers.openai.com/api/docs/pricing, checked 2026-10-23.

Rates applied per 1M tokens: input $1.75, cache read $0.175, cache write $1.75, output $14.

Cheapest comparable models for this exact volume Price difference only; test quality and latency before switching.

Same provider (OpenAI), comparable on paper: GPT-6 Luna $59.01 · GPT-5.6 Luna $128 · GPT-5.4 mini $481 · GPT-6.1 Sol $1,144 · GPT-6 Sol $1,180

Price history for this model: 2026-10-23: GPT-5.2 added to the table: $1.75 input / $14 output per 1M tokens.

GPT-5.4 mini OpenAI batch

Your rows: "gpt-5-4-mini". Price source: developers.openai.com/api/docs/pricing, checked 2026-10-23.

Rates applied per 1M tokens: input $0.375, cache read $0.0375, cache write $0.375, output $2.25.

Cheapest comparable models for this exact volume Price difference only; test quality and latency before switching.

Same provider (OpenAI), comparable on paper: GPT-6 Luna $60.88 · GPT-5.6 Luna $133 · GPT-6.1 Sol $1,218 · GPT-6 Sol $1,218 · GPT-5.6 Terra $1,330

Price history for this model: 2026-10-17: GPT-5.4 mini first recorded: $0.75 input / $4.5 output per 1M tokens.

GPT-4o mini OpenAI standard deprecated

Your rows: "gpt-4o-mini-2024-07-18" (matched without the date suffix). Price source: developers.openai.com/api/docs/pricing, checked 2026-10-23.

Rates applied per 1M tokens: input $0.15, cache read $0.075, cache write $0.15, output $0.6.

Cheapest comparable models for this exact volume Price difference only; test quality and latency before switching.

Same provider (OpenAI), comparable on paper: GPT-6 Luna $217 · GPT-5.6 Luna $473 · GPT-5.4 mini $1,773 · GPT-6.1 Sol $4,328 · GPT-6 Sol $4,342

Price history for this model: 2026-10-17: GPT-4o mini first recorded: $0.15 input / $0.6 output per 1M tokens.

Every model in our table, priced on your volume

"Comparable on paper" means: still offered, a context window at least as large as your current model's (up to 400K), image input if yours takes images, and tool calling if yours has it. It says nothing about output quality. Price difference only; test quality and latency before switching.

GPT-4o: all 66 other models for your volume

GPT-5.2: all 66 other models for your volume

GPT-5.4 mini: all 66 other models for your volume

GPT-4o mini: all 66 other models for your volume

Method and sources

• Prices: tensorust.com/llm/, open data at tensorust.com/data/v1/llm/models.json (CC BY 4.0). Each model's price is read from its provider's pricing page on the date shown.
• Cached input is billed at the model's cache-read price; cache writes at its shortest-lifetime cache-write price; batch traffic at its batch prices; flex, priority and off-peak rows at those tier prices. Long-context prices apply when your average prompt per request is over the model's threshold (needs a request count).
• Alternatives are priced on the same token counts. Different models tokenize text differently, so the same text can be 10 to 30% more or fewer tokens on another provider.
• Scenarios (batch, cache, flex) are what-ifs on your volume, not predictions.