aicostbudget-ai/ai-api-pricing

LLM API Pricing Dataset

Prices per 1M tokens for about 100 LLM and AI API models from OpenAI, Anthropic, Google, xAI, DeepSeek, Mistral, Cohere, Moonshot AI, Alibaba Qwen, Meta, and other providers. These are examples, not an exhaustive provider list. Every row links to the official provider pricing page it was verified against and carries its own last_verified_at date. Prices are updated as providers change them.

Rows	One row per public model (about 100)
Prices	Input, cached input, and output per 1M tokens in USD, plus non-token units such as pages, characters, and images
Beyond list prices	Cache, batch, long-context tiers, multimodal, and time-of-day pricing in prices.json / prices.csv
Evidence	Official source URL and verification date on every row; unknown prices stay empty, never zero
License	CC BY 4.0

Estimate your own bill: compare two models with your token counts and success rates in the Model Switch Cost Calculator, or see the cheapest LLM APIs right now.

• Live dataset and documentation
• Source repository
• Fixed v1.1.0 release (snapshot 2026-09-30)
• Version DOI
• Methodology
• Price history: dated snapshots and verified price changes
• LLM API Pricing Calculator Space and the LLM API Pricing & Cost collection

Prices change frequently. Verify production billing decisions against the provider pricing page, contract, billing dashboard, and invoice.

Quick start

import pandas as pd

url = (
    "https://huggingface.co/datasets/"
    "aicostbudget-ai/ai-api-pricing/resolve/main/train.csv"
)

pricing = pd.read_csv(url)

# Ten cheapest token-priced models by output price
token_rows = pricing.dropna(subset=["output_price_per_1m_tokens"])
print(
    token_rows.sort_values("output_price_per_1m_tokens")[
        ["provider", "model", "input_price_per_1m_tokens", "output_price_per_1m_tokens"]
    ].head(10)
)

With Hugging Face Datasets:

from datasets import load_dataset

pricing = load_dataset(
    "aicostbudget-ai/ai-api-pricing",
    split="train",
)

print(pricing)
print(pricing[0])

The train split is the current published pricing snapshot. It is not intended as model-training supervision data.

Use it from code or an AI assistant

• JavaScript / TypeScript: npm install ai-api-pricing looks up prices, lists models, and estimates or compares workload costs. Every returned price includes its official source URL and verification date. (npm package)
• AI assistants (MCP): npx -y ai-api-pricing-mcp runs a read-only MCP server with tools for model prices, filtered model lists, monthly cost estimates, comparisons, and recent price changes. It is listed in the official MCP Registry as io.github.aicostbudget/ai-api-pricing-mcp and on Glama. (npm package)

Both packages read the GitHub Pages /api/v1 compatibility projection described in the technical notes below. Source code: packages/.

Compare pricing with your workload

The Model Switch Cost Calculator compares API costs for two models based on your input and output tokens, monthly request volume, and assumed success rates. It estimates monthly spending, cost per successful task, and the break-even success rate for a model switch.

The results depend on your inputs and assumptions. The tool does not benchmark model quality, measure actual success rates, automate routing, or guarantee savings.

To measure success rates on your own models, use the open LLM cost-per-task benchmark: 16 tasks with code-only validators, run with your own API keys and priced from this dataset.

Published files

• train.csv: a lightweight CSV projection consumed by the Hugging Face Dataset Viewer
• prices.csv: the complete normalized CSV download
• prices.json: metadata plus the complete public Website record set
• meta.json: generation time, verification date, and coverage counts for this export

Download them from the Hugging Face files, or use the live dataset's public JSON and public CSV downloads.

https://huggingface.co/datasets/aicostbudget-ai/ai-api-pricing/resolve/main/train.csv
https://huggingface.co/datasets/aicostbudget-ai/ai-api-pricing/resolve/main/prices.csv
https://huggingface.co/datasets/aicostbudget-ai/ai-api-pricing/resolve/main/prices.json
https://huggingface.co/datasets/aicostbudget-ai/ai-api-pricing/resolve/main/meta.json

train.csv lists the most-searched providers first and intentionally omits the large serialized pricing_tiers_json, time_pricing_json, pricing_components_json, conditional_usage_allowances_json, model_selection_json, access_evidence_json, and binding_evidence_json columns for Dataset Viewer usability. It preserves every exported model and all other CSV fields. Complete tier, time, component, conditional allowance, transport, alternative-measurement, optional-feature, model-selection, and access-evidence details remain available in prices.csv and prices.json. Schema 1.8.0 keeps eligibility-scoped short-term allowances separate from standard paid usage tiers.

Unknown or unavailable prices are null in JSON and empty in CSV; they are never rewritten as zero.

Hugging Face generates the Dataset Viewer and auto-converted Parquet from train.csv as convenience views. They are not an independent pricing source. Viewer and search indexing can lag behind the raw published artifacts, so use prices.json, prices.csv, and meta.json for the corresponding published version when displayed counts or dates differ.

Fields

Field	Meaning
provider_id	Stable provider identifier.
provider	Human-readable provider name.
model_id	Stable public model identifier.
model	Human-readable model name.
input_price_per_1m_tokens	Standard input-token price per 1M tokens.
cached_input_price_per_1m_tokens	Compatibility/default cached-input or read-like scalar when available; provider cache semantics can differ.
output_price_per_1m_tokens	Standard output-token price per 1M tokens.
currency	Pricing currency; current records use USD.
pricing_unit	Normalized unit; current records use 1M tokens.
status	Public lifecycle/status label from the Website projection or marked fallback.
availability	Public availability label.
official_source_url	Provider source used by the verified pricing pipeline.
verification_status	Canonical record status such as verified, review_required, or partially_verified.
last_verified_at	Record verification date; it is not refreshed merely because an artifact is rebuilt.
checked_at	Date the source was checked, kept separate from verification.
effective_from	Provider-stated effective date when available.
effective_until	End date when a selected price record has one.
notes	Source context, projection warning, or legacy fallback warning.
pricing_tier_count	Number of structured context-dependent pricing tiers.
pricing_tiers	JSON-only structured tier records, including thresholds and source references.
pricing_tiers_json	CSV-only lossless JSON serialization of pricing_tiers; [] when no tiers apply.
pricing_components	JSON-only full conditional pricing representation: component, decimal-string amount, conditions, effective range, source IDs and URLs, and verification status; [] when none apply.
pricing_components_json	Final CSV column containing the same pricing_components array as compact deterministic JSON; [] when none apply.

The scalar input, cached-input, and output fields remain the compatibility/default pricing view. They are not deprecated. pricing_components represents conditions and units that one scalar cannot: context thresholds, batch and processing modes, cache writes, multimodal usage, tiered pricing, and non-token services. The pricing contract defines the full structure. For structured pricing details, see the repository's model schema.

Access Policy V1 adds optional, independent access and identity-binding metadata. JSON exposes access_status, access_checked_at, official access_evidence, binding_status, and binding_evidence. CSV appends five columns after all existing columns: access_status, access_checked_at, access_evidence_json, binding_status, and binding_evidence_json. Consumers using named columns remain compatible; consumers assuming a fixed column count must accept these additions. Access verification timestamps are separate from price freshness. Access restrictions do not filter this factual pricing export. Older pinned exports without access metadata retain their original JSON and CSV contract when reproduced.

Freshness semantics

• generated_at is the artifact generation timestamp copied from the checked-in pricing metadata.
• last_verified_at belongs to each record and means the price was verified against its source.
• checked_at records a source check that may not have reached verified status.
• review_required and partially_verified records keep last_verified_at empty; neither checked_at nor generated_at is substituted.
• Rebuilding this distribution does not rewrite record verification dates.
• Validation requires generated_at >= max(last_verified_at) and generated_at >= max(checked_at), and rejects future timestamps.

For current coverage and timestamps, inspect meta.json. The source repository also contains schema checks, price validation, history, snapshots, and freshness workflows.

Technical notes: how this export is built

This Hugging Face dataset is the machine-readable distribution of the public AI API pricing records published by AICostBudget. It is not a separately curated subset: train.csv, prices.csv, and prices.json are generated from the same Pricing V2 public projection used by the AICostBudget Dataset page and download APIs. The Hugging Face dataset mirrors that Pricing V2 public projection; GitHub Pages /api/v1 is a separate compatibility projection and may have a different record count by design.

The export is deterministic:

1. Start from the checked-in Pricing V2 Website projection.
2. Exclude compatibility aliases because they are not independent public pricing rows.
3. Exclude rows marked excluded_default_candidate by the verified pricing pipeline.
4. Preserve Website legacy fallback rows only when the public Pricing Table adapter still exposes them, including the warning that they are not default-safe verified prices.
5. Export exactly the resulting public Website key set; no provider, family, or status-specific allowlist is maintained for Hugging Face.

The canonical provider and pricing facts live in the AICostBudget pricing data repository. Website-owned legacy display metadata is used only for explicitly marked fallback rows. The exporter does not fetch or guess prices.

Methodology and contributions

• Methodology
• Contributing
• Data license
• Code license

Dataset data is licensed under CC BY 4.0; repository code is licensed under MIT. The front matter license applies to the dataset data.

Prices are accepted from official provider pricing pages, documentation, APIs, or announcements. Third-party aggregators and search snippets are not final price evidence.

Citation

For reproducible analysis, cite the frozen v1.1.0 release and its version DOI (snapshot 2026-09-30). The live dataset and Hugging Face main can change after that snapshot; for latest operational use, cite the live dataset with your access date instead. This is a version DOI, not a claim that it identifies the latest dataset.

AICostBudget. AICostBudget AI API Pricing Dataset, v1.1.0, snapshot 2026-09-30.
Version DOI: 10.5281/zenodo.23087250
DOI: https://doi.org/10.5281/zenodo.23087250
GitHub Release: https://github.com/aicostbudget/ai-api-pricing-data/releases/tag/v1.1.0
Canonical live dataset: https://aicostbudget.com/en/datasets/ai-api-pricing

@misc{aicostbudget_2026_v110,
  author = {{AICostBudget}},
  title = {AICostBudget AI API Pricing Dataset},
  year = {2026},
  version = {1.1.0},
  doi = {10.5281/zenodo.23087250},
  url = {https://doi.org/10.5281/zenodo.23087250},
  note = {Snapshot 2026-09-30}
}

For the changing dataset, record the access date and inspect meta.json for export generation and aggregate verification timestamps. Record-level verification remains in each row's last_verified_at; do not replace it with the access date or generated_at.

Disclaimer

This dataset is informational. Prices can change, provider conditions vary, and verification timestamps may lag provider changes. Check official provider documentation before production billing decisions. AICostBudget is independent and is not affiliated with or endorsed by the listed providers. Provider names and trademarks belong to their respective owners.

If this dataset saves you time, consider liking it on Hugging Face so you can find future pricing updates more easily.