Which of the big three closed-model APIs has got cheaper, and by how much? We compared OpenAI, Anthropic and Google from primary evidence: monthly Internet Archive captures of each provider’s own pricing page. OpenAI’s captures start in January 2023, Anthropic’s in March 2024 and Google’s Gemini API in April 2024, all running to November 2026. Every price below links the capture it comes from. BroadRouter is BroadNet’s website. BroadNet does not sell these closed models and is not part of this comparison.
The data is our LLM API price history dataset. For today’s prices next to open-weight models, see open-weight vs closed model API cost.
How did GPT, Claude and Gemini prices compare over time?
The newest model in each line at four points in time, at its list price in the latest capture on or before that date. USD per 1M tokens, input / output. Each price links its archive capture.
| Line | End of June 2024 | End of June 2025 | End of June 2026 | Early November 2026 |
|---|---|---|---|---|
| OpenAI, GPT main line | gpt-4o: 5.00 / 15.00 | gpt-4.1: 2.00 / 8.00 | gpt-5.5: 5.00 / 30.00 | gpt-6.1-sol: 2.00 / 10.00 |
| OpenAI, small line (mini, Luna) | gpt-3.5-turbo-0125: 0.50 / 1.50 | gpt-4.1-mini: 0.40 / 1.60 | gpt-5.4-mini: 0.75 / 4.50 | gpt-6-luna: 0.10 / 0.50 |
| Anthropic, Opus | Claude Opus 3: 15.00 / 75.00 | Claude Opus 4: 15.00 / 75.00 | Claude Opus 4.8: 5.00 / 25.00 | Claude Opus 5.5: 4.00 / 20.00 |
| Anthropic, Sonnet | Claude 3 Sonnet: 3.00 / 15.00 | Claude Sonnet 4: 3.00 / 15.00 | Claude Sonnet 4.6: 3.00 / 15.00 | Claude Sonnet 5.5: 2.00 / 10.00 |
| Anthropic, Haiku | Claude Haiku 3: 0.25 / 1.25 | Claude Haiku 3.5: 0.80 / 4.00 | Claude Haiku 4.5: 1.00 / 5.00 | Claude Haiku 4.5: 1.00 / 5.00 |
| Google, Gemini Pro | Gemini 1.5 Pro: 3.50 / 10.50 | Gemini 2.5 Pro Preview: 1.25 / 10.00 | Gemini 3.1 Pro Preview: 2.00 / 12.00 | Gemini 4 Argon: 2.00 / 10.00 (promotional) |
| Google, Gemini Flash | Gemini 1.5 Flash: 0.35 / 1.05 | Gemini 2.5 Flash Preview: 0.15 / 0.60 | Gemini 3.5 Flash: 1.50 / 9.00 | Gemini 3.8 Flash: 0.75 / 3.75 (promotional) |
“Newest” means the model in that line first captured most recently. Lines are grouped by the providers’ own model names; the grouping is ours. Gemini 4 Argon and Gemini 3.8 Flash carry promotional labels on the page (“through November 30, 2026” and “through December 31, 2026”).
How much have the three providers repriced existing models?
Hardly at all. A model’s price on these three pages almost never changes after it is listed:
| Provider | Monthly captures used | Coverage | Models seen in 2+ months | Repriced | Cuts | Rises |
|---|---|---|---|---|---|---|
| OpenAI | 48 | Jan 2023 to Nov 2026 | 77 | 5 | 5 | 0 |
| Anthropic | 39 | Mar 2024 to Nov 2026 | 26 | 1 | 1 | 1 |
| Google (Gemini API) | 32 | Apr 2024 to Nov 2026 | 23 | 2 | 3 | 0 |
The repricings were OpenAI’s o3 (−80%), gpt-4o, gpt-5.6-sol, o1-mini and gpt-3.5-turbo; Google’s Gemini 1.5 Pro (twice) and Gemini 1.5 Flash; and Anthropic’s Claude 3.5 Haiku, which rose and then fell. Details and archive links are in LLM API price cuts 2023 to 2026. Everything else in the tables below is the price of a new version.
OpenAI: how did GPT prices change?
The main GPT line, first archived price of each model:
| Model | First captured | Price then, in / out | Latest capture | Price then, in / out |
|---|---|---|---|---|
| gpt-4 (8K context) | 1 Apr 2023 | 30.00 / 60.00 | 5 Nov 2023 | 30.00 / 60.00 |
| gpt-4-1106-preview | 1 Dec 2023 | 10.00 / 30.00 | 4 Mar 2026 | 10.00 / 30.00 |
| gpt-4o | 1 Jun 2024 | 5.00 / 15.00 | 4 Mar 2026 | 2.50 / 10.00 |
| gpt-4.1 | 3 May 2025 | 2.00 / 8.00 | 4 Mar 2026 | 2.00 / 8.00 |
| gpt-5 | 2 Sep 2025 | 1.25 / 10.00 | 4 Mar 2026 | 1.25 / 10.00 |
| gpt-5.2 | 2 Jan 2026 | 1.75 / 14.00 | 4 Mar 2026 | 1.75 / 14.00 |
| gpt-5.4 | 1 Apr 2026 | 2.50 / 15.00 | 1 Aug 2026 | 2.50 / 15.00 |
| gpt-5.5 | 3 May 2026 | 5.00 / 30.00 | 1 Aug 2026 | 5.00 / 30.00 |
| gpt-5.6-sol | 1 Aug 2026 | 5.00 / 30.00 | 1 Nov 2026 | 4.00 / 20.00 |
| gpt-6.1-sol | 1 Oct 2026 | 2.00 / 10.00 | 1 Nov 2026 | 2.00 / 10.00 |
From GPT-4 to GPT-6.1 Sol the blended price (3 input tokens per output token) fell 9.4 times. But 2026 was a climb before the drop: GPT-5 at USD 1.25 / USD 10.00 (September 2025 capture) was followed by GPT-5.2, GPT-5.4 and then GPT-5.5 at USD 5.00 / USD 30.00, 3.3 times GPT-5 blended. gpt-5.6-sol was cut to USD 4.00 / USD 20.00 a month after its first capture, and GPT-6.1 Sol arrived at USD 2.00 / USD 10.00.
OpenAI’s small line:
| Model | First captured | Price then, in / out | Latest capture | Price then, in / out |
|---|---|---|---|---|
| gpt-3.5-turbo-0125 | 1 Feb 2024 | 0.50 / 1.50 | 4 Mar 2026 | 0.50 / 1.50 |
| GPT-4o mini | 1 Aug 2024 | 0.15 / 0.60 | 1 Apr 2025 | 0.15 / 0.60 |
| gpt-4.1-mini | 3 May 2025 | 0.40 / 1.60 | 4 Mar 2026 | 0.40 / 1.60 |
| gpt-5-mini | 2 Sep 2025 | 0.25 / 2.00 | 4 Mar 2026 | 0.25 / 2.00 |
| gpt-5.4-mini | 1 Apr 2026 | 0.75 / 4.50 | 1 Aug 2026 | 0.75 / 4.50 |
| gpt-5.6-luna | 1 Aug 2026 | 0.20 / 1.20 | 1 Sep 2026 | 0.20 / 1.20 |
| gpt-6-luna | 1 Oct 2026 | 0.10 / 0.50 | 1 Nov 2026 | 0.10 / 0.50 |
GPT-6 Luna, at USD 0.10 / USD 0.50, is 24% below GPT-4o mini blended. In between, gpt-5.4-mini reached USD 0.75 / USD 4.50.
Anthropic: how did Claude prices change?
Claude Opus:
| Model | First captured | Price then, in / out | Latest capture | Price then, in / out |
|---|---|---|---|---|
| Claude Opus 3 | 4 Mar 2024 | 15.00 / 75.00 | 4 May 2026 | 15.00 / 75.00 |
| Claude Opus 4 | 3 Jun 2025 | 15.00 / 75.00 | 1 Nov 2026 | 15.00 / 75.00 |
| Claude Opus 4.1 | 2 Sep 2025 | 15.00 / 75.00 | 1 Nov 2026 | 15.00 / 75.00 |
| Claude Opus 4.5 | 1 Dec 2025 | 5.00 / 25.00 | 1 Nov 2026 | 5.00 / 25.00 |
| Claude Opus 4.6 | 1 Mar 2026 | 5.00 / 25.00 | 1 Nov 2026 | 5.00 / 25.00 |
| Claude Opus 4.7 | 4 May 2026 | 5.00 / 25.00 | 1 Nov 2026 | 5.00 / 25.00 |
| Claude Opus 4.8 | 1 Jun 2026 | 5.00 / 25.00 | 1 Nov 2026 | 5.00 / 25.00 |
| Claude Opus 5 | 1 Aug 2026 | 5.00 / 25.00 | 1 Nov 2026 | 5.00 / 25.00 |
| Claude Opus 5.5 | 2 Oct 2026 | 4.00 / 20.00 | 1 Nov 2026 | 4.00 / 20.00 |
Claude Sonnet:
| Model | First captured | Price then, in / out | Latest capture | Price then, in / out |
|---|---|---|---|---|
| Claude 3 Sonnet | 4 Mar 2024 | 3.00 / 15.00 | 1 Jan 2025 | 3.00 / 15.00 |
| Claude Sonnet 3.5 | 1 Jul 2024 | 3.00 / 15.00 | 2 Oct 2025 | 3.00 / 15.00 |
| Claude Sonnet 3.7 | 3 Mar 2025 | 3.00 / 15.00 | 4 May 2026 | 3.00 / 15.00 |
| Claude Sonnet 4 | 3 Jun 2025 | 3.00 / 15.00 | 1 Nov 2026 | 3.00 / 15.00 |
| Claude Sonnet 4.5 | 2 Oct 2025 | 3.00 / 15.00 | 1 Nov 2026 | 3.00 / 15.00 |
| Claude Sonnet 4.6 | 1 Mar 2026 | 3.00 / 15.00 | 1 Nov 2026 | 3.00 / 15.00 |
| Claude Sonnet 5 | 1 Jul 2026 | 2.00 / 10.00 (promotional) | 1 Nov 2026 | 2.00 / 10.00 |
| Claude Sonnet 5.5 | 2 Oct 2026 | 2.00 / 10.00 | 2 Oct 2026 | 2.00 / 10.00 |
Claude Haiku:
| Model | First captured | Price then, in / out | Latest capture | Price then, in / out |
|---|---|---|---|---|
| Claude Haiku 3 | 4 Mar 2024 | 0.25 / 1.25 | 4 May 2026 | 0.25 / 1.25 |
| Claude Haiku 3.5 | 2 Nov 2024 | 0.25 / 1.25 | 1 Nov 2026 | 0.80 / 4.00 |
| Claude Haiku 4.5 | 19 Nov 2025 | 1.00 / 5.00 | 1 Nov 2026 | 1.00 / 5.00 |
Three distinct stories:
- Opus fell in two steps: USD 15 / USD 75 for three versions, USD 5 / USD 25 from Opus 4.5, USD 4 / USD 20 for Opus 5.5. That is 3.75 times lower blended.
- Sonnet held still for six versions at USD 3 / USD 15, from Claude 3 Sonnet (March 2024 capture) to Sonnet 4.6 (March 2026). Sonnet 5’s first capture was labelled introductory (“through August 31, 2026”), but later captures show the same USD 2 / USD 10 with no label.
- Haiku rose. Haiku 4.5 at USD 1 / USD 5 is 4 times Haiku 3. The Claude 3.5 Haiku card moved from USD 0.25 / USD 1.25 (2 November 2024 capture) to USD 1 / USD 5 (3 December 2024) and USD 0.80 / USD 4 (1 January 2025).
Google: how did Gemini prices change?
Gemini Pro:
| Model | First captured | Price then, in / out | Latest capture | Price then, in / out |
|---|---|---|---|---|
| Gemini 1.0 Pro | 2 Apr 2024 | 0.50 / 1.50 | 1 Jan 2025 | 0.50 / 1.50 |
| Gemini 1.5 Pro | 2 Apr 2024 | 7.00 / 21.00 | 2 Oct 2025 | 1.25 / 5.00 |
| Gemini 2.5 Pro Preview | 2 May 2025 | 1.25 / 10.00 | 3 Jun 2025 | 1.25 / 10.00 |
| Gemini 2.5 Pro | 1 Jul 2025 | 1.25 / 10.00 | 1 Nov 2026 | 1.25 / 10.00 |
| Gemini 3 Pro Preview | 6 Dec 2025 | 2.00 / 12.00 | 1 Mar 2026 | 2.00 / 12.00 |
| Gemini 3.1 Pro Preview | 1 Mar 2026 | 2.00 / 12.00 | 1 Nov 2026 | 2.00 / 12.00 |
| Gemini 4 Argon | 1 Nov 2026 | 2.00 / 10.00 (promotional) | 1 Nov 2026 | 2.00 / 10.00 (promotional) |
Gemini Flash:
| Model | First captured | Price then, in / out | Latest capture | Price then, in / out |
|---|---|---|---|---|
| Gemini 1.5 Flash | 3 Jun 2024 | 0.35 / 1.05 | 2 Oct 2025 | 0.075 / 0.30 |
| Gemini 2.0 Flash | 11 Feb 2025 | 0.10 / 0.40 | 1 Aug 2026 | 0.10 / 0.40 |
| Gemini 2.5 Flash Preview | 2 May 2025 | 0.15 / 0.60 | 3 Jun 2025 | 0.15 / 0.60 |
| Gemini 2.5 Flash | 1 Jul 2025 | 0.30 / 2.50 | 1 Nov 2026 | 0.30 / 2.50 |
| Gemini 3 Flash Preview | 1 Jan 2026 | 0.50 / 3.00 | 1 Nov 2026 | 0.50 / 3.00 |
| Gemini 3.5 Flash | 2 Jun 2026 | 1.50 / 9.00 | 1 Oct 2026 | 1.50 / 9.00 |
| Gemini 3.6 Flash | 1 Aug 2026 | 1.50 / 7.50 | 1 Nov 2026 | 0.75 / 3.75 (promotional) |
| Gemini 3.7 Flash | 1 Sep 2026 | 0.75 / 3.75 (promotional) | 1 Oct 2026 | 0.75 / 3.75 (promotional) |
| Gemini 3.8 Flash | 1 Oct 2026 | 0.75 / 3.75 (promotional) | 1 Nov 2026 | 0.75 / 3.75 (promotional) |
Google’s two lines moved in opposite directions:
- Pro got much cheaper, then a bit dearer. Gemini 1.5 Pro was cut twice in 2024, from USD 7.00 / USD 21.00 to USD 1.25 / USD 5.00 (short-context prices). Gemini 3 and 3.1 Pro Preview list USD 2.00 / USD 12.00, about twice 1.5 Pro’s final price blended.
- Flash got much dearer. After its 2024 cut, Gemini 1.5 Flash cost USD 0.075 / USD 0.30. Gemini 3.5 Flash was first captured at USD 1.50 / USD 9.00, about 26 times that blended. The current Gemini 3.8 Flash, at a price the page marks as running through 31 December 2026, is still 8.6 times Gemini 2.0 Flash blended.
What does this mean for choosing a model?
- Expect cheaper prices from new versions, not from waiting. All three providers leave old models at their old price. To capture a lower price you must migrate and retest.
- Do not assume small models get cheaper. At Anthropic and Google, the newest small model in late 2026 costs more than the 2024 one did. OpenAI’s newest small model is the exception.
- Read the labels. Several current prices at Google and one first capture at Anthropic were marked promotional or introductory. Budget on the price after the promotion.
- Compare across vendors at each tier. Today’s per-token prices for these models, next to open-weight models, are in open-weight vs closed model API cost and the LLM API price index.
Methodology
- Source. Monthly Internet Archive captures of each provider’s own pricing pages (OpenAI’s four pricing URLs over time, Anthropic’s pricing and docs pages, Google’s ai.google.dev pricing pages). Each price keeps its archive URL and a verbatim quote. Dataset and method: /data/llm-api-price-history/; all rows: CSV.
- Which price. Standard pay-as-you-go text input and output per 1M tokens; short-context price where the page splits by prompt length. Batch, flex and priority tiers excluded. Early Gemini prices per 1,000 characters were not converted and are not shown.
- Promotions. Prices the page labels as introductory or time-limited are marked “promotional” and not counted as cuts or rises.
- Lines. We grouped models into lines by the providers’ names (GPT main line, mini and Luna, Opus, Sonnet, Haiku, Gemini Pro, Gemini Flash). The grouping and the side-by-side table are computed by our script from the dataset JSON; “first captured” is the earliest capture listing the model, not its launch date.
- Ratios. Blended = 3 input tokens per output token. For example GPT-4 (8K) blended is (3 × 30 + 60) ÷ 4 = 37.5 and GPT-6.1 Sol is 4.0, a ratio of 9.4.
Where BroadNet Inference fits
BroadRouter is BroadNet’s own site. BroadNet Inference sells open-weight models only (DeepSeek V4.1 Flash, GLM 5.3, Qwen3.8 27B and BroadNet’s build of Gemma 4), not GPT, Claude or Gemini, and has no archived price history. For comparison only: GLM 5.3 on BroadNet is $0.78 input and $3.46 output per 1M tokens, and DeepSeek V4.1 Flash $0.175 and $0.52. Whether an open model can replace a closed one depends on your tasks; open-weight vs closed models for business covers the trade-offs.
