Model pricing source audit
Verified on 2026-10-09 against official provider documentation. Prices below are USD per million tokens, written as input / output. The Python registry stores these amounts divided by 1,000,000. This is a standard, uncached text API snapshot, not a live availability check or a guarantee of access for an account.
The supported-price tables below match the project's selected built-in registry
in src/longguard/pricing.py. A model can be excluded by project selection even
while its provider still offers it. The decision tables distinguish those
exclusions from retirement or unverifiable pricing. Use list_supported_models()
for the current supported IDs and explicit per-token overrides for excluded models.
OpenAI
The official pricing page and individual model cards verify standard text rates for the selected entries.
| Model ID | Input / output | Source |
|---|---|---|
gpt-6-astra |
10 / 50 | Astra |
gpt-6.1-sol |
2 / 10 | 6.1 Sol |
gpt-6-sol |
2 / 10 | Sol |
gpt-6-luna |
0.10 / 0.50 | Luna |
gpt-5.6, gpt-5.6-sol |
4 / 20 | 5.6 Sol and its alias |
gpt-5.6-terra |
2 / 12 | Terra |
gpt-5.6-luna |
0.20 / 1.20 | 5.6 Luna |
gpt-5.5 |
5 / 30 | 5.5 |
gpt-5.4 |
2.50 / 15 | 5.4 |
gpt-5.4-mini |
0.75 / 4.50 | 5.4 Mini |
gpt-5.4-nano |
0.20 / 1.25 | 5.4 Nano |
gpt-5.2 |
1.75 / 14 | 5.2 |
gpt-5.1 |
1.25 / 10 | 5.1 |
gpt-5 |
1.25 / 10 | 5 |
gpt-5-mini |
0.25 / 2 | 5 Mini |
gpt-5-nano |
0.05 / 0.40 | 5 Nano |
gpt-4.1, gpt-4.1-2025-04-14 |
2 / 8 | 4.1 |
gpt-4.1-mini |
0.40 / 1.60 | 4.1 Mini |
gpt-4.1-nano |
0.10 / 0.40 | 4.1 Nano |
gpt-4o, gpt-4o-2024-08-06, gpt-4o-2024-11-20 |
2.50 / 10 | 4o |
gpt-4o-mini, gpt-4o-mini-2024-07-18 |
0.15 / 0.60 | 4o Mini |
GPT-6, GPT-5.6, GPT-5.5 and GPT-5.4 base rates apply through 272,000 input tokens; larger requests have 2x input and 1.5x output rates. GPT-5.4 Mini/Nano use their own model-card prices. GPT-5.6 Sol's promotional price is guaranteed at least through November 21, 2026. Cache writes, cached inputs, service tiers, regions and tool use incur separate rates. See the linked cards and pricing details.
Provider retirement dates follow the authoritative deprecation schedule, which takes precedence over a model card that still displays historical prices:
| Existing IDs | Decision and reason |
|---|---|
gpt-4-turbo-preview |
Remove; its underlying January 2024 preview retired March 26, 2026. |
o1-mini |
Remove; retired October 27, 2025. |
o3 |
Excluded by project selection; its snapshot's scheduled provider retirement is December 11, 2026. |
gpt-4-turbo, gpt-4, gpt-3.5-turbo |
Excluded by project selection; scheduled provider retirement is October 23, 2026. |
o1, o3-mini, o4-mini |
Excluded by project selection; this does not assert provider retirement. |
gpt-4o, gpt-4o-mini, gpt-4o-mini-2024-07-18 |
Retain their verified rates. |
The added GPT-5/5 Mini/5 Nano releases retire December 11, 2026; GPT-5.1 and GPT-5.4 Nano retire April 1, 2027. Their presence estimates currently billable usage, and is not a recommendation to choose a deprecated model. Only listed snapshots resolve; no arbitrary suffix removal or provider-prefix stripping is performed. GPT-4o's May 2024 snapshot is intentionally not mapped to the cheaper August/November 2024 rates.
Anthropic
The pricing table and model lifecycle table support the following public Claude API entries:
| Model ID | Input / output |
|---|---|
claude-fable-5-1, claude-fable-5 |
10 / 50 |
claude-opus-5-5 |
4 / 20 |
claude-opus-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6 |
5 / 25 |
claude-opus-4-5, claude-opus-4-5-20251101 |
5 / 25 |
claude-sonnet-5-5, claude-sonnet-5 |
2 / 10 |
claude-sonnet-4-6 |
3 / 15 |
claude-sonnet-4-5, claude-sonnet-4-5-20250929 |
3 / 15 |
claude-haiku-5-5 |
0.10 / 0.50 |
claude-haiku-4-5, claude-haiku-4-5-20251001 |
1 / 5 |
Haiku 5.5's base rate applies through 100,000 input tokens; larger prompts cost 0.50 / 2.50. Sonnet 4.5 is deprecated but available through November 30, 2026. Restricted-access Mythos models are omitted. Standard prices exclude caching, batch, fast mode, tools, and geography premiums. See the pricing details.
Aliases are verified in the official Opus 4.5, Sonnet 4.5, and Haiku 4.5 cards. From 4.6 onwards, dateless IDs are pinned releases; see model IDs and versioning.
Existing table audit, using the official retirement history:
| Existing IDs | Decision and reason |
|---|---|
claude-3-5-sonnet, claude-3-5-sonnet-20241022 |
Remove; Sonnet 3.5 retired October 28, 2025. |
claude-3-5-haiku, claude-3-5-haiku-20241022 |
Remove; Haiku 3.5 retired February 19, 2026. |
claude-3-opus, claude-3-opus-20240229 |
Remove; Opus 3 retired January 5, 2026. |
claude-3-sonnet |
Remove; Sonnet 3 retired July 21, 2025. |
claude-3-haiku, claude-3-haiku-20240307 |
Remove; Haiku 3 retired April 20, 2026. |
claude-sonnet-4, claude-opus-4 |
Remove; their dated releases retired June 15, 2026. |
claude-haiku-4 |
Remove; no such public model ID verified in the current catalog. |
Google Gemini Developer API
The standard paid pricing table verifies these text rates:
| Model ID | Input / output |
|---|---|
gemini-3.8-flash, gemini-3.7-flash, gemini-3.6-flash |
0.75 / 3.75 |
gemini-3.5-flash |
1.50 / 9 |
gemini-3.5-flash-lite |
0.30 / 2.50 |
gemini-3.1-flash-lite |
0.25 / 1.50 |
gemini-3.1-pro-preview, gemini-3.1-pro-preview-customtools |
2 / 12 |
gemini-3-flash-preview |
0.50 / 3 |
Scheduled price change: the 3.8, 3.7, and 3.6 Flash rates above expire December 31, 2026; they become 1.50 / 7.50 on January 1, 2027. Pro rates apply through 200,000 input tokens: larger 3.1 Pro prompts cost 4 / 18. Output includes thinking tokens. Audio input, caching, batch/flex/priority, and tools have different rates. Pricing.
Gemini 2.5 models are excluded from the project's built-in registry. At the verification date, the deprecation table restricts them to existing users without an announced shutdown; their exclusion does not assert that the provider has retired them.
| Existing ID | Decision and reason |
|---|---|
gemini-1.5-pro, gemini-1.5-flash, gemini-1.5-flash-8b |
Remove; shut down September 29, 2025, per release notes. |
gemini-2.0-flash, gemini-2.0-flash-lite |
Remove; shut down June 1, 2026, per deprecations. |
gemini-2.5-pro, gemini-2.5-flash, gemini-2.5-flash-lite |
Excluded by project selection. |
Groq
The current model catalog verifies:
| Model ID | Input / output | Availability |
|---|---|---|
openai/gpt-oss-120b |
0.15 / 0.60 | Production |
openai/gpt-oss-20b |
0.075 / 0.30 | Production |
qwen/qwen3.8-27b |
0.80 / 4 | Preview |
These are Groq-hosted prices, not OpenAI API or generic open-weight-model prices. Speech and safety-only models are outside this text-generation scope.
| Existing ID | Decision and reason |
|---|---|
llama-3.1-8b-instant, llama-3.3-70b-versatile |
Remove fixed prices; free/developer shutdown August 16, 2026, while committed-spend enterprise access remains with contact-sales pricing. |
llama-3.1-70b-versatile |
Remove; shut down January 24, 2025. |
llama-3.1-405b-reasoning |
Remove; neither current availability nor its listed price was verified in the official Groq catalog. |
Shutdown dates and enterprise exceptions come from Groq deprecations; custom-price availability comes from the catalog.
Mistral
The standard USD pricing table and individual cards verify:
| Model ID | Input / output | Model card |
|---|---|---|
mistral-large-4, mistral-large-4-0 |
0.68 / 2.09 | Large 4 |
mistral-large-2512, mistral-large-latest |
0.50 / 1.50 | Large 3 |
mistral-medium-3-5, mistral-medium-latest |
1.50 / 7.50 | Medium 3.5 |
mistral-small-2603, mistral-small-latest |
0.15 / 0.60 | Small 4 |
ministral-14b-2512, ministral-14b-latest |
0.20 / 0.20 | 14B |
ministral-8b-2512, ministral-8b-latest |
0.15 / 0.15 | 8B |
ministral-3b-2512, ministral-3b-latest |
0.10 / 0.10 | 3B |
codestral-2508, codestral-latest |
0.30 / 0.90 | Codestral |
Short-lived launch sale: Large 4's normal rate is 1.36 / 4.18.
The October 6 release advertises
50% off for two weeks; refresh the snapshot around October 20, 2026. Large 4
is a public preview. Its -4-0 ID is also used in official
reasoning examples.
The -latest mappings are explicit in the
Vibe CLI configuration documentation.
In particular, mistral-large-latest still refers to Large 3, not the Large 4
preview. Aliases can change prices and behavior under the
lifecycle policy.
| Existing ID | Decision and reason |
|---|---|
mistral-large |
Remove; the bare identifier's current API mapping was not verified. Replace with the verified IDs above. |
mistral-small |
Remove; historical endpoint renamed in 2024, and current mapping not verified. Use mistral-small-latest or mistral-small-2603. |
mistral-nemo |
Remove; bare ID and price not verified in current catalog. Historical API ID was open-mistral-nemo. |
codestral |
Remove; bare ID's API mapping not verified. Use codestral-latest or codestral-2508; current rates differ from the old table. |
The historical Small/Nemo names come from the official changelog. These removals identify unsupported or unverifiable registry entries, not a claim that every old Mistral release has retired on every hosting platform.
Cohere
The live-model catalog and model cards verify the selected entries:
| Model ID | Input / output | Source |
|---|---|---|
command-a-03-2025 |
2.50 / 10 | Command A |
command-r-08-2024 |
0.15 / 0.60 | Command R |
command-r7b-12-2024 |
0.0375 / 0.15 | Command R7B |
| Excluded IDs | Decision and reason |
|---|---|
command-r, command-r-03-2024, command-r-plus, command-r-plus-04-2024 |
Excluded by project selection; the generic aliases target older March/April 2024 snapshots. |
command-r-plus-08-2024 |
Excluded by project selection; this does not assert provider retirement. |
The older snapshots are deprecated for new users, while existing-user access continues until retirement under Cohere's lifecycle definition. See the legacy pricing FAQ for historical pricing.
The live command-a-plus-05-2026, command-a-reasoning-08-2025,
command-a-translate-08-2025, and command-a-vision-07-2025 have no verified
flat paid token rate. Their cards state free API access through rate limits,
with production through Model Vault or sales. Do not invent a paid rate or
represent contracted production as zero cost. Sources:
A+,
Reasoning,
Translate,
Vision.
Estimation limits
The current ModelPricing interface has only input/output prices. It cannot
represent long-context tiers, cache hits/writes, region or service-tier
multipliers, modality rates, server-tool charges, or contract pricing. Built-in
entries therefore estimate ordinary uncached text usage at the base tier.
Supply explicit per-token overrides for other billing arrangements. Account
permissions and provider lifecycle changes can affect whether an API call works.