Result first · exact-model workload
AI API model cost leaderboard
Choose an exact model, enter your monthly token volume, and see every fully comparable provider cost before reading the method.
Exact model · openai/gpt-oss-120b
GPT-OSS 120B monthly cost comparison
Enter your monthly token workload and compare the exact openai/gpt-oss-120b route across every provider with complete matching evidence.
| Rank status | Provider | Input price | Cache-read price | Output price | Monthly cost | Exact configuration | Sources and freshness |
|---|---|---|---|---|---|---|---|
| —Evidence is aging or stale | Fireworks AIaggregator | $0.15 / 1M input tokens | $0.015 / 1M cached input tokens | $0.6 / 1M output tokens | UnavailableEvidence is aging or stale | accounts/fireworks/models/gpt-oss-120b · serverless standard · chat completions |
|
| —Evidence is aging or stale | OpenRouterrouter | $0.037 / 1M input tokens | $0.03 / 1M cached input tokens | $0.17 / 1M output tokens | UnavailableEvidence is aging or stale | openai/gpt-oss-120b · standard · chat completions |
|
| —Evidence is aging or stale | Together AIaggregator | $0.15 / 1M input tokens | Not published | $0.6 / 1M output tokens | UnavailableEvidence is aging or stale | openai/gpt-oss-120b · serverless · chat completions |
|
Missing and excluded providers (6)
- Fireworks AI: Evidence is aging or stale.
- OpenRouter: Evidence is aging or stale.
- Together AI: Evidence is aging or stale.
- OpenAI direct: No accepted owner-hosted GPT-OSS 120B API price is on file; the open model release is not treated as a hosted API offer.
- DeepInfra: No accepted exact-route price observation was found in this review.
- Novita AI: No accepted exact-route price observation was found in this review.
Strict boundary: Shared hosted routes only. Dedicated capacity, throughput, latency, rate limits, retries, taxes, support, and service quality are outside this price order. Together has no accepted cached-input meter, so it is excluded when cached usage is greater than zero.
Embeddable result
Put this exact-model comparison on your site
The SVG card is cookie-free and links back to the same source-auditable workload. It shows a rank only while at least three routes qualify.
<script async referrerpolicy="no-referrer" src="https://apidir.dev/embed/model-cost.js?model=gpt-oss-120b&input=100000000&cached=0&output=20000000"></script>
Exact-model alert · controlled rollout
Join the GPT-OSS 120B price alert waitlist.
Confirm your address to join the waitlist. Delivery remains paused until sender, suppression, and evidence checks are ready; ambiguous or stale changes stay quarantined.
Exact models only
Switch model without mixing families.
Qwen3.8 Max is not replaced by open-weight Qwen3.8, and broad Claude, Kimi, Grok, or Qwen family labels never enter a cost table.
openai/gpt-oss-120b
GPT-OSS 120B
Enter your monthly token workload and compare the exact openai/gpt-oss-120b route across every provider with complete matching evidence.
Compare monthly workload →Evidence incompleteqwen3.8-max
Qwen3.8 Max
Compare the hosted Qwen3.8 Max service without mixing in the feature-different open-weight Qwen3.8 model.
Compare monthly workload →Evidence incompletekimi-k3
Kimi K3
Compare the exact kimi-k3 route across owner-direct, router, and serverless channels while keeping provider configuration visible.
Compare monthly workload →Evidence incompletegrok-4.6
Grok 4.6
Two exact Grok 4.6 channels are visible, but APIDir does not announce a cheapest provider until a third fully comparable route passes review.
Compare monthly workload →Evidence incompleteclaude-sonnet-5
Claude Sonnet 5
Two exact Claude Sonnet 5 channels are visible, but APIDir does not announce a cheapest provider until a third fully comparable route passes review.
Compare monthly workload →Evidence incompletemistral-medium-3-5-26-04
Mistral Medium 3.5
Compare the versioned Mistral Medium 3.5 release across owner-direct and routed access while keeping upstream identity and missing cache evidence visible.
Compare monthly workload →Configuration-gatedExact meter scopes
Runway Gen-4.5 API pricing
Compare the exact Gen-4.5 text-to-video output-second meter without mixing Runway subscriptions, Gen-4 Turbo, or unrelated hosted video models.
Inspect evidence →Configuration-gatedExact meter scopes
Luma Ray 3.2 API pricing
See the result first for Ray 3.2 text-to-video at 720p, five seconds, SDR, no references, no loop, and no HDR or EXR export.
Inspect evidence →Configuration-gatedExact meter scopes
Hailuo 2.3 API pricing
Compare one exact Hailuo 2.3 standard text-to-video request at 768p and six seconds without mixing fast, pro, image-to-video, or 1080p offers.
Inspect evidence →Configuration-gatedExact meter scopes
Wan 2.7 API pricing
Compare exact Wan 2.7 text-to-video routes at 720p with no audio input, excluding image, reference, edit, 1080p, and promotional top-up pricing.
Inspect evidence →Configuration-gatedExact meter scopes
Recraft V3 API pricing
Compare one Recraft V3 raster text-to-image output across three provider channels while keeping vector generation and image-editing operations outside the order.
Inspect evidence →Configuration-gatedExact meter scopes
Seedream 4.5 API pricing
Compare the exact Seedream 4.5 base text-to-image route at one 2K output without mixing edits, sequential generation, 4K settings, or newer Seedream models.
Inspect evidence →Configuration-gatedExact meter scopes
AI video API pricing ranking by exact configuration
APIDir publishes a winner only when at least three fresh provider offers pass the same reviewed model, meter, mode, resolution, duration, and audio scope.
Inspect evidence →Evidence inventoryCoverage, not quality
AI API pricing evidence coverage ranking
Count fresh accepted price observations by provider without turning source coverage into a reliability, quality, or recommendation score.
Inspect coverage →Method after the answer
The comparison behaves like a calibrated scale.
The model ID is the object on the scale. Input, cache-read, and output meters are three separate weights. If one provider is missing a required weight, APIDir removes that route from the calculation instead of guessing.
- Lock one exact provider route for one exact model.
- Require accepted first-party provider evidence and a current fresh-until time.
- Normalize every used token meter to USD per one million tokens.
- Calculate monthly cost from the submitted uncached input, cached input, and output volumes.
- Publish ranks only at three or more complete providers. Preserve ties.
Questions developers ask
When does APIDir announce the lowest-cost provider?
Only when at least three unique providers have complete, accepted, fresh evidence for the exact model and every token meter used by the workload. Two providers are shown without rank; one becomes a price card.
Why can changing cached tokens remove a provider?
A route without a reviewed cached-input price cannot price cached usage. It remains comparable when cached usage is zero, but APIDir excludes it instead of substituting the uncached input rate.
Do affiliate links affect the table?
No. Eligibility, cost calculation, ties, and ordering use only reviewed price evidence. Any future affiliate link must be labelled and can never change the result.