---
title: "LLM API price cuts 2023–2026, with archive sources | BroadRouter"
description: "The largest LLM API price cuts from 2023 to 2026, each with before and after Internet Archive links: o3, Codestral, Gemini 1.5, plus the rises few people noticed."
url: https://broadrouter.com/blog/llm-api-price-cuts-2023-2026/
lang: en
updated: 2026-11-15
publisher: BroadRouter (a BroadNet site); API by BroadNet Inference, https://inference.broadnet.ai
---

# LLM API price cuts 2023–2026: the biggest drops, with sources

By the BroadNet Inference team · 8 min read · Published 2026-11-15 · Updated 2026-11-16

The largest list-price cuts in our archive of 11 providers' pricing pages were 80% on OpenAI o3 (mid-2025) and Mistral Codestral (late 2024), 75% on Gemini 1.5 Flash and 81% on Mistral Small 3 at Together AI. Makers' cuts were large (median 58%); hosts' were small (median 14%). We also found 52 rises, including Claude 3.5 Haiku.

## Key takeaways

-   We found 100 list-price cuts on the same model across 11 providers from 2023 to 2026; 12 of them halved the blended price or more.
-   Model makers cut rarely but deeply: 15 cuts with a median of 58%. Hosts cut often but lightly: 85 cuts with a median of 14%.
-   Every cut below links the archived page before and after, so you can check the price yourself.
-   Prices also rose 52 times, among them Claude 3.5 Haiku (USD 0.25 to USD 1.00 input between two captures in late 2024) and DeepSeek V4 Pro in 2026.
-   Dates are Internet Archive capture dates. A cut happened somewhere between the two captures, usually about a month apart.

**On this page**

1.  How many LLM API price cuts were there?
2.  What were the biggest cuts by model makers?
3.  What were the biggest cuts by hosts?
4.  Which price rises do fewer people know about?
5.  What should you take from the cuts?
6.  Methodology
7.  Where BroadNet Inference fits

LLM API prices are often said to fall fast. That is true for new models, but how large were the cuts to existing models, and can you prove them? We built this list from primary evidence: Internet Archive captures of 11 providers’ own pricing pages, one per month, from January 2023 to November 2026. Each row below links the archived page before and after the change. BroadRouter is BroadNet’s website; BroadNet is not in the archive data.

The full set of 3,332 archived prices is our [LLM API price history dataset](https://broadrouter.com/data/llm-api-price-history/). For how often prices move at all, see [how often LLM API prices change](https://broadrouter.com/blog/how-often-llm-api-prices-change/).

## How many LLM API price cuts were there?

We found 100 list-price cuts on the same model: no price went up and at least one went down between two consecutive captures. The median cut was 19% on a blended price (3 input tokens per output token). 12 of the 100 halved the blended price or more.

The cuts split into two very different groups:

| Who cut | Cuts | Median cut (blended) |
| --- | --- | --- |
| Model makers (OpenAI, Anthropic, Google, DeepSeek, Mistral, Cohere; xAI made none) | 15 | −58% |
| Hosts (DeepInfra, Together AI, Groq, Fireworks AI) | 85 | −14% |

## What were the biggest cuts by model makers?

All 15 cuts on makers’ own pricing pages, largest first. Prices are USD per 1M tokens, input / output. A gap note means the two captures are more than 60 days apart, so the cut could have happened at any point in that window.

| Provider and model | Captures | Before, in / out | After, in / out | Blended change | Archive |
| --- | --- | --- | --- | --- | --- |
| Mistral AI Codestral | 7 Sep 2024 → 6 Oct 2024 | 1.00 / 3.00 | 0.20 / 0.60 | −80% | [before](https://web.archive.org/web/20240907034418id_/https://mistral.ai/technology/), [after](https://web.archive.org/web/20241006123931id_/https://mistral.ai/technology/) |
| OpenAI o3 | 2 Jun 2025 → 2 Jul 2025 | 10.00 / 40.00 | 2.00 / 8.00 | −80% | [before](https://web.archive.org/web/20250602152701id_/https://openai.com/api/pricing/), [after](https://web.archive.org/web/20250702152237id_/https://openai.com/api/pricing/) |
| Google (Gemini API) Gemini 1.5 Flash | 3 Aug 2024 → 4 Sep 2024 | 0.35 / 1.05 | 0.075 / 0.30 | −75% | [before](https://web.archive.org/web/20240803102002id_/https://ai.google.dev/pricing), [after](https://web.archive.org/web/20240904171757id_/https://ai.google.dev/pricing) |
| DeepSeek deepseek-reasoner | 1 Sep 2025 → 4 Oct 2025 | 0.55 / 2.19 | 0.28 / 0.42 | −67% | [before](https://web.archive.org/web/20250901020532id_/https://api-docs.deepseek.com/quick_start/pricing), [after](https://web.archive.org/web/20251004112357id_/https://api-docs.deepseek.com/quick_start/pricing) |
| Cohere Command R | 15 Aug 2024 → 13 Sep 2024 | 0.50 / 1.50 | 0.15 / 0.60 | −65% | [before](https://web.archive.org/web/20240815122219id_/https://cohere.com/pricing), [after](https://web.archive.org/web/20240913064105id_/https://cohere.com/pricing) |
| OpenAI o1-mini | 9 Jan 2025 → 2 Jan 2026 (358 days apart) | 3.00 / 12.00 | 1.10 / 4.40 | −63% | [before](https://web.archive.org/web/20250109100914id_/https://openai.com/api/pricing/), [after](https://web.archive.org/web/20260102061741id_/https://platform.openai.com/docs/pricing) |
| OpenAI gpt-3.5-turbo | 3 Jun 2023 → 2 Jan 2026 (944 days apart) | 2.00 / 2.00 | 0.50 / 1.50 | −62% | [before](https://web.archive.org/web/20230603020027id_/https://openai.com/pricing), [after](https://web.archive.org/web/20260102061741id_/https://platform.openai.com/docs/pricing) |
| Google (Gemini API) Gemini 1.5 Pro | 1 Oct 2024 → 2 Nov 2024 | 3.50 / 10.50 | 1.25 / 5.00 | −58% | [before](https://web.archive.org/web/20241001092414id_/https://ai.google.dev/pricing), [after](https://web.archive.org/web/20241102172747id_/https://ai.google.dev/pricing) |
| Google (Gemini API) Gemini 1.5 Pro | 2 May 2024 → 3 Jun 2024 | 7.00 / 21.00 | 3.50 / 10.50 | −50% | [before](https://web.archive.org/web/20240502061735id_/https://ai.google.dev/pricing), [after](https://web.archive.org/web/20240603135204id_/https://ai.google.dev/pricing) |
| Mistral AI Mistral Large | 5 May 2024 → 25 May 2025 (385 days apart) | 4.00 / 12.00 | 2.00 / 6.00 | −50% | [before](https://web.archive.org/web/20240505230813id_/https://mistral.ai/technology/), [after](https://web.archive.org/web/20250525061708id_/https://mistral.ai/pricing) |
| OpenAI gpt-4o | 1 Oct 2024 → 12 Nov 2024 | 5.00 / 15.00 | 2.50 / 10.00 | −42% | [before](https://web.archive.org/web/20241001064825id_/https://openai.com/api/pricing/), [after](https://web.archive.org/web/20241112050536id_/https://openai.com/api/pricing/) |
| Mistral AI Mistral Large 2 | 7 Sep 2024 → 6 Oct 2024 | 3.00 / 9.00 | 2.00 / 6.00 | −33% | [before](https://web.archive.org/web/20240907034418id_/https://mistral.ai/technology/), [after](https://web.archive.org/web/20241006123931id_/https://mistral.ai/technology/) |
| OpenAI gpt-5.6-sol | 1 Aug 2026 → 1 Sep 2026 | 5.00 / 30.00 | 4.00 / 20.00 | −29% | [before](https://web.archive.org/web/20260801091023id_/https://developers.openai.com/api/docs/pricing), [after](https://web.archive.org/web/20260901103152id_/https://developers.openai.com/api/docs/pricing) |
| Cohere Command R+ | 15 Aug 2024 → 13 Sep 2024 | 3.00 / 15.00 | 2.50 / 10.00 | −27% | [before](https://web.archive.org/web/20240815122219id_/https://cohere.com/pricing), [after](https://web.archive.org/web/20240913064105id_/https://cohere.com/pricing) |
| Anthropic Claude 3.5 Haiku | 3 Dec 2024 → 1 Jan 2025 | 1.00 / 5.00 | 0.80 / 4.00 | −20% | [before](https://web.archive.org/web/20241203010242id_/https://www.anthropic.com/pricing), [after](https://web.archive.org/web/20250101035229id_/https://www.anthropic.com/pricing) |

Notes on the table:

-   **Gemini 1.5 Pro was cut twice**, from USD 7.00 / USD 21.00 to USD 1.25 / USD 5.00 in 2024 captures: 79% lower blended overall. These are the short-context prices; the page listed higher rates for long prompts.
-   **gpt-3.5-turbo and o1-mini** compare captures far apart: the later one is OpenAI’s pricing page in January 2026, and the archive cannot say when in between the price changed. By then the gpt-3.5-turbo name may also point to a newer snapshot.
-   **Claude 3.5 Haiku’s 20% cut** came one month after a 300% rise on the same pricing card. See the rises section below.
-   **OpenAI gpt-5.6-sol** moved from USD 5.00 / USD 30.00 to USD 4.00 / USD 20.00 between the August and September 2026 captures.

## What were the biggest cuts by hosts?

The 10 largest of the 85 host cuts. Hosts serve open-weight models made by others, so a cut here usually follows new hardware, a quantised variant or competition on the same model.

| Provider and model | Captures | Before, in / out | After, in / out | Blended change | Archive |
| --- | --- | --- | --- | --- | --- |
| Together AI Mistral Small 3 | 1 Jan 2026 → 10 Feb 2026 | 0.80 / 0.80 | 0.10 / 0.30 | −81% | [before](https://web.archive.org/web/20260101184249id_/https://www.together.ai/pricing), [after](https://web.archive.org/web/20260210173839id_/https://www.together.ai/pricing) |
| DeepInfra Qwen3-235B-A22B-Instruct-2507 | 3 Feb 2026 → 4 Mar 2026 | 0.071 / 0.463 | 0.071 / 0.10 | −54% | [before](https://web.archive.org/web/20260203123011id_/https://deepinfra.com/pricing), [after](https://web.archive.org/web/20260304081141id_/https://deepinfra.com/pricing) |
| Groq Llama 4 Maverick (17Bx128E) | 5 Apr 2025 → 3 May 2025 | 0.50 / 0.77 | 0.20 / 0.60 | −47% | [before](https://web.archive.org/web/20250405230317id_/https://groq.com/pricing/), [after](https://web.archive.org/web/20250503095534id_/https://groq.com/pricing/) |
| Fireworks AI DeepSeek R1 | 5 Feb 2025 → 4 Mar 2025 | 8.00 / 8.00 | 3.00 / 8.00 | −47% | [before](https://web.archive.org/web/20250205030147id_/https://fireworks.ai/pricing), [after](https://web.archive.org/web/20250304004954id_/https://fireworks.ai/pricing) |
| DeepInfra OpenChat-3.5 | 17 Apr 2024 → 20 May 2024 | 0.13 / 0.13 | 0.07 / 0.07 | −46% | [before](https://web.archive.org/web/20240417184927id_/https://deepinfra.com/pricing), [after](https://web.archive.org/web/20240520235826id_/https://deepinfra.com/pricing) |
| DeepInfra Llama-2-7b-chat | 17 Apr 2024 → 20 May 2024 | 0.13 / 0.13 | 0.07 / 0.07 | −46% | [before](https://web.archive.org/web/20240417184927id_/https://deepinfra.com/pricing), [after](https://web.archive.org/web/20240520235826id_/https://deepinfra.com/pricing) |
| DeepInfra Mistral-7B | 17 Apr 2024 → 20 May 2024 | 0.13 / 0.13 | 0.07 / 0.07 | −46% | [before](https://web.archive.org/web/20240417184927id_/https://deepinfra.com/pricing), [after](https://web.archive.org/web/20240520235826id_/https://deepinfra.com/pricing) |
| DeepInfra Lzlv-70b | 1 Aug 2024 → 19 Sep 2024 | 0.59 / 0.79 | 0.35 / 0.40 | −43% | [before](https://web.archive.org/web/20240801121125id_/https://deepinfra.com/pricing), [after](https://web.archive.org/web/20240919152346id_/https://deepinfra.com/pricing) |
| DeepInfra Llama-2-13b-chat | 17 Apr 2024 → 20 May 2024 | 0.22 / 0.22 | 0.13 / 0.13 | −41% | [before](https://web.archive.org/web/20240417184927id_/https://deepinfra.com/pricing), [after](https://web.archive.org/web/20240520235826id_/https://deepinfra.com/pricing) |
| DeepInfra MythoMax-L2-13b | 17 Apr 2024 → 20 May 2024 | 0.22 / 0.22 | 0.13 / 0.13 | −41% | [before](https://web.archive.org/web/20240417184927id_/https://deepinfra.com/pricing), [after](https://web.archive.org/web/20240520235826id_/https://deepinfra.com/pricing) |

Together AI’s 81% cut on Mistral Small 3 is the largest single cut in the dataset. Many of DeepInfra’s cuts came in batches: several models changed between the same two captures, as in April to May 2024.

## Which price rises do fewer people know about?

Cuts get the headlines, but we also recorded 52 list-price rises on the same model. A few stand out:

-   **Claude 3.5 Haiku, Anthropic.** The [2 November 2024 capture](https://web.archive.org/web/20241102082318id_/https://www.anthropic.com/pricing) of Anthropic’s pricing page carries a Claude 3.5 Haiku card at USD 0.25 input and USD 1.25 output per 1M tokens, the same as Claude 3 Haiku. The [3 December 2024 capture](https://web.archive.org/web/20241203010242id_/https://www.anthropic.com/pricing) shows USD 1 and USD 5, four times higher. The [1 January 2025 capture](https://web.archive.org/web/20250101035229id_/https://www.anthropic.com/pricing) shows USD 0.80 and USD 4. We report what the captures show; we do not say when or why Anthropic set each price.
-   **Codestral, Mistral.** After the 80% cut in late 2024, codestral-latest went from USD 0.2 / USD 0.6 to USD 0.3 / USD 0.9 between the [2 January 2025](https://web.archive.org/web/20250102174557id_/https://mistral.ai/technology/) and [3 February 2025](https://web.archive.org/web/20250203181801id_/https://mistral.ai/technology/) captures, a 50% rise.
-   **DeepSeek V4 Pro, DeepSeek.** USD 0.435 input and USD 0.87 output in the [1 August 2026 capture](https://web.archive.org/web/20260801170952id_/https://api-docs.deepseek.com/quick_start/pricing/); USD 1.32 and USD 3.96 at peak in the [3 October 2026 capture](https://web.archive.org/web/20261003073808id_/https://api-docs.deepseek.com/quick_start/pricing/), which also lists an off-peak rate at half.
-   **DeepInfra and Together AI.** 40 of the 52 rises were at DeepInfra and 7 at Together AI, for example Together’s Gemma 4 31B from USD 0.20 / USD 0.50 to USD 0.39 / USD 0.97 between May and June 2026.

Every rise, with its archive links, is in [Do LLM API prices only go down?](https://broadrouter.com/blog/llm-api-price-increases/)

## What should you take from the cuts?

-   **Big cuts come from makers, and they are rare.** 15 maker cuts in nearly four years, from 6 of the 7 makers. Do not plan on your model getting cheaper.
-   **Cheaper usually means a newer model.** Most of the fall in flagship prices came from new versions at lower prices. See [OpenAI vs Anthropic vs Google API price history](https://broadrouter.com/blog/openai-anthropic-google-api-price-history/).
-   **Host prices drift both ways.** Hosts cut 85 times and raised prices 48 times. If you use an open model, compare hosts regularly: [the same model costs different amounts at different hosts](https://broadrouter.com/blog/same-llm-different-host-prices/).
-   **Check the evidence before you quote a cut.** The archive links above let you confirm each price on the provider’s own page as it was.

## Methodology

-   **Source.** Internet Archive captures of each provider’s own pricing page, one per provider per month from 2023, read by per-provider extraction scripts. 308 of 417 fetched captures had prices. Every row keeps its archive URL and a verbatim quote. Dataset page: [/data/llm-api-price-history/](https://broadrouter.com/data/llm-api-price-history/). Raw rows: [CSV](https://broadrouter.com/data/llm-api-price-history.csv).
-   **What counts as a cut.** Two consecutive captures of the same model (same API ID or name) where no price went up and at least one went down. Standard pay-as-you-go text prices only; batch, flex and priority tiers are excluded. Where a page splits by context length, the short-context price is used.
-   **Excluded.** Moves into or out of a price the page marks as promotional or introductory (2 moves), one row with a zero price on one side only (a probable page error), and euro prices.
-   **Percentages.** Blended change uses 3 input tokens per output token. Medians are over the 15 maker cuts and 85 host cuts. Our script recomputes every figure from the dataset JSON with the same rules as the dataset page.
-   **Dates.** Capture dates, not announcement dates. The median gap between the two captures of a change is 32 days.

## Where BroadNet Inference fits

BroadRouter is BroadNet’s own site. BroadNet Inference has no archive history, so it is not in these tables. It publishes current prices in a machine-readable [pricing.json](https://inference.broadnet.ai/pricing.json?utm_source=broadrouter&utm_medium=referral&utm_campaign=blog-llm-api-price-cuts-2023-2026&utm_content=body), for example DeepSeek V4.1 Flash at $0.175 input and $0.52 output per 1M tokens. Other hosts list the same model for less on some lines; [compare hosts](https://broadrouter.com/blog/same-llm-different-host-prices/) before you choose.

Free tools for this

-   [LLM API cost calculator](https://broadrouter.com/tools/llm-api-cost-calculator/) Price a workload on every model at once, with cache hits, reasoning and a flat-plan check.
-   [Flat-rate vs per-token LLM cost calculator](https://broadrouter.com/tools/flat-rate-vs-per-token-calculator/) Break-even between an unlimited plan and paying per token, per agent.
-   [Prompt caching savings calculator](https://broadrouter.com/tools/prompt-caching-calculator/) Bill with and without caching, by prefix share and hit rate.
-   [Prompt token diff](https://broadrouter.com/tools/prompt-token-diff/) Two prompt versions side by side: changed lines, exact token delta and the monthly cost difference.

[All free tools](https://broadrouter.com/tools/)

## Terms in this article

[Token](https://broadrouter.com/glossary/token/)

A token is the unit of text a language model reads and writes: a word, part of a word, a punctuation mark or a byte sequence produced by the model's tokenizer. LLM APIs count, limit and usually bill usage in tokens.

[Input and output tokens](https://broadrouter.com/glossary/input-and-output-tokens/)

Input tokens are the tokens a language model reads in a request, including the system prompt, conversation history, tool definitions and images. Output tokens are the tokens it generates in reply, including any reasoning. LLM APIs usually price them separately, and output usually costs more.

[Open-weight model](https://broadrouter.com/glossary/open-weight-model/)

An open-weight model is a machine learning model whose trained weights are published for anyone to download and run, usually under a licence with conditions. It differs from fully open-source AI, which also releases the training code and detailed information about the training data, so others can study and reproduce the model.

[Context window](https://broadrouter.com/glossary/context-window/)

A context window is the maximum number of tokens a language model can work with in a single request. It covers the prompt, conversation history, tool definitions, images and the generated output, including any reasoning tokens. Anything that does not fit must be cut, summarised or split across several requests.

[Pay-as-you-go](https://broadrouter.com/glossary/pay-as-you-go/)

Pay-as-you-go is a pricing model in which you pay only for what you use, metered per unit such as tokens, requests or audio minutes, with no subscription or minimum commitment. Most LLM APIs bill this way, from a prepaid balance or a periodic invoice.

## Questions

### What was the biggest LLM API price cut?

In our archive, the largest cut on a model maker's own page was 80%, shared by OpenAI o3 (USD 10.00 / USD 40.00 to USD 2.00 / USD 8.00 between captures on 2 June and 2 July 2025) and Mistral Codestral (USD 1.00 / USD 3.00 to USD 0.20 / USD 0.60 between 7 September and 6 October 2024). The largest overall was 81% on Mistral Small 3 at Together AI.

### How much did Gemini 1.5 Pro fall in price?

Two cuts in 2024 took Gemini 1.5 Pro from USD 7.00 input and USD 21.00 output per 1M tokens (April and May 2024 captures, short-context price) to USD 1.25 and USD 5.00 (November 2024 capture), 79% lower blended.

### Do these dates show when the provider announced the cut?

No. They are the dates of the Internet Archive captures before and after the change. The new price took effect at some point between them. We make no claim about announcement dates.

### Did any LLM API prices go up?

Yes, 52 times in our archive, with a median rise of 40%. Most were at DeepInfra and Together AI, but makers raised prices too, for example Mistral Codestral in early 2025 and DeepSeek V4 Pro in 2026.

### Why is BroadNet not in the table?

The Internet Archive had no captures of BroadNet Inference's pricing pages when we checked on 15 November 2026, so there is no BroadNet price history to compare.

## Keep reading

-   [How often do LLM API prices change? 3 years of archived pricing pages](https://broadrouter.com/blog/how-often-llm-api-prices-change/)
-   [Do LLM API prices only go down? 52 rises and how to budget for them](https://broadrouter.com/blog/llm-api-price-increases/)
-   [OpenAI vs Anthropic vs Google API price history, 2023–2026](https://broadrouter.com/blog/openai-anthropic-google-api-price-history/)
-   [Same model, different host: how much LLM API prices vary](https://broadrouter.com/blog/same-llm-different-host-prices/)
-   [LLM API pricing explained: input, cached, output and reasoning](https://broadrouter.com/blog/llm-api-pricing-explained/)
-   [LLM API price changes in October 2026: 3 rises, no cuts](https://broadrouter.com/blog/llm-api-price-changes-october-2026/)
-   [How to reduce LLM API costs: a 10-step checklist, in order](https://broadrouter.com/blog/how-to-reduce-llm-api-costs/)

Try it on your real workload. New accounts start with $5 of credit for Gemma and transcription. DeepSeek, GLM and Qwen3.8 need a top-up.

[Get started](https://inference.broadnet.ai/account/?utm_source=broadrouter&utm_medium=referral&utm_campaign=blog-llm-api-price-cuts-2023-2026&utm_content=page-cta) ·

**Sources**

-   [BroadRouter dataset: LLM API price history from archived provider pricing pages (checked 2026-11-15)](https://broadrouter.com/data/llm-api-price-history/)
-   [BroadRouter dataset: LLM API price history, CSV download (checked 2026-11-15)](https://broadrouter.com/data/llm-api-price-history.csv)
-   [Internet Archive capture of openai.com/api/pricing, 2 June 2025: o3 at USD 10.00 / USD 40.00 (checked 2026-11-15)](https://web.archive.org/web/20250602152701id_/https://openai.com/api/pricing/)
-   [Internet Archive capture of openai.com/api/pricing, 2 July 2025: o3 at USD 2.00 / USD 8.00 (checked 2026-11-15)](https://web.archive.org/web/20250702152237id_/https://openai.com/api/pricing/)
-   [Internet Archive capture of mistral.ai/technology, 6 October 2024: Codestral at USD 0.2 / USD 0.6 (checked 2026-11-15)](https://web.archive.org/web/20241006123931id_/https://mistral.ai/technology/)
-   [Internet Archive capture of ai.google.dev/pricing, 4 September 2024: Gemini 1.5 Flash (checked 2026-11-15)](https://web.archive.org/web/20240904171757id_/https://ai.google.dev/pricing)
-   [Internet Archive capture of anthropic.com/pricing, 2 November 2024: Claude 3.5 Haiku card (checked 2026-11-15)](https://web.archive.org/web/20241102082318id_/https://www.anthropic.com/pricing)
-   [Internet Archive capture of anthropic.com/pricing, 3 December 2024: Claude 3.5 Haiku card (checked 2026-11-15)](https://web.archive.org/web/20241203010242id_/https://www.anthropic.com/pricing)
