# OpenAI and Anthropic Slash LLM Prices by Up to 80%

> Published 2026-10-26 · https://www.promptzone.com/neha_wu/openai-and-anthropic-slash-llm-prices-by-up-to-80-2do0

OpenAI and Anthropic have executed a series of drastic price cuts for their flagship LLM APIs, with reductions of up to **80%** occurring within just a few months. The trend, [highlighted in a recent Hacker News discussion](https://minimaxir.com/2026/10/cheap-llms/), signals a fierce new phase of competition focused on commoditizing core model access.

> **Model:** GPT-4o | **Input Price:** $0.15 → $0.03 per 1M tokens | **Output Price:** $0.60 → $0.15 per 1M tokens
> **Model:** Claude 3.5 Sonnet | **Input Price:** $3.00 → $0.60 per 1M tokens | **Output Price:** $15.00 → $3.00 per 1M tokens

## What Happened: A Pricing Freefall

In October 2024, OpenAI set the stage by cutting GPT-4 Turbo's input price by **67%**. Anthropic responded by slashing Claude 3.5 Sonnet's prices by **80%** for input and output in early 2025. OpenAI countered weeks later with its own **80% output cost reduction** for GPT-4o.

The net result is that generating 1 million tokens of text from GPT-4o now costs **$150**, down from **$600**. For Claude 3.5 Sonnet, the same output now costs **$3,000**, down from **$15,000**.


![OpenAI and Anthropic Slash LLM Prices by Up to 80%](https://promptzone-community.s3.amazonaws.com/uploads/articles/33q9a5gcok5r25o014im.jpeg)

## The New Price Landscape

The cuts have reshaped the cost-per-performance matrix. Claude 3.5 Sonnet, once prohibitively expensive for high-volume tasks, is now in a closer race with GPT-4o on price, though it remains more expensive for long outputs.

| Feature | GPT-4o (New) | Claude 3.5 Sonnet (New) | GPT-4 Turbo (Pre-Oct '24) |
|---------|--------------|--------------------------|----------------------------|
| Input / 1M tokens | **$0.03** | **$0.60** | $0.01 |
| Output / 1M tokens | **$0.15** | **$3.00** | $0.03 |
| Context Window | 128K | 200K | 128K |
| Key Strength | Speed, Vision | Long-context reasoning, Coding | Was the price leader |

Early testers on Hacker News report the primary impact is on **agentic workflows and long-form generation**, where output tokens dominate cost. A task that previously cost $10 may now cost $2.

## Why the Sudden Cuts? Scale and Competition

The price war is driven by three concrete factors. First, **inference optimization** has drastically reduced the computational cost per token for model providers. Techniques like speculative decoding and improved hardware utilization are yielding real savings.

Second, the emergence of capable **open-weight models** like Meta's Llama 3.1 405B and Mistral's offerings has created a competitive ceiling. While these models require self-hosting, their existence pressures commercial API pricing.

Third, **market saturation** for simple chat completion is pushing providers to compete on cost for developers building scalable applications. The goal is to lock in the next generation of AI-native apps before they choose an alternative.

{% details "Technical Context: The Cost of Inference" %}
Inference cost is primarily a function of FLOPs (floating-point operations) per token and infrastructure efficiency. A 2024 estimate from SemiAnalysis suggested the *raw compute cost* for generating a million tokens with a model like GPT-4 could be under $0.01 at scale. The gap between raw cost and API price covers R&D, profit, and support. Aggressive competition compresses this margin.
{% enddetails %}

## How to Try the New Prices

Accessing the new rates requires no model changes, only a billing update. For existing users, the lower prices are automatically applied to your API consumption.

- **OpenAI Platform:** Use the [official OpenAI API](https://platform.openai.com/docs/models) with your API key. The `gpt-4o` model name already reflects the new pricing.
- **Anthropic Console:** The reduced prices for `claude-3-5-sonnet-20241022` are live in the [Anthropic API](https://docs.anthropic.com/en/docs/about-claude/models).
- **Monitoring:** Use the cost tracking tools in both platforms' dashboards to recalculate your monthly forecasts. A project that burned through a $500 budget last month might now only use $100.

## Alternatives and Comparisons

The price cuts force a reevaluation of the entire API landscape. Google's Gemini 1.5 Flash and Pro models, along with startups like Cohere, are now under significant pressure to match the new cost baseline.

| Provider | Model | Input Price / 1M Tokens | Best For |
|----------|-------|--------------------------|----------|
| **OpenAI** | GPT-4o | **$0.03** | General purpose, fastest iteration |
| **Anthropic** | Claude 3.5 Sonnet | $0.60 | Long documents, complex instruction following |
| **Google** | Gemini 1.5 Flash | $0.075 | Cheap, high-volume preprocessing |
| **Mistral AI** | Mistral Large 2 | $0.50 | European data residency, strong multilingual |
| **Open-Weight** | Llama 3.1 405B | Variable (~$1-4/hr for compute) | Full control, no data sharing |

For teams with strict data governance or very predictable, high-volume workloads, **self-hosting a model like Llama 3.1 70B** on cloud instances may now offer a better total cost of ownership than just months ago.

## Who Should Use This Now

**Act immediately if you are:** a startup scaling an AI feature with variable, output-heavy traffic; a researcher processing large corpora with API models; or a developer who previously found Claude 3.5 Sonnet too expensive for long-form tasks.

**Re-evaluate your stack if you are:** heavily invested in Gemini Flash for cost reasons, as its price advantage has eroded. Also, consider if cheaper API costs change the calculus on building versus buying for specific tasks.

**Stick with your current plan if you are:** a hobbyist with minimal usage (costs were already low), or an enterprise with a committed spend discount or a workflow locked to a specific model's unique capabilities.

> **Bottom line:** The cost of intelligence as a utility has dropped precipitously, making ambitious, token-heavy applications financially viable for many more developers.

The race to the bottom on price is a direct subsidy for innovation. It shifts the competitive moat from who can afford API calls to who can build the most compelling application logic on top of them. Expect the next wave of cuts to target multimodal input or specialized endpoints, as providers seek new dimensions to compete on beyond the commodity of text.