# GPT-4.5 ![Organization](https://img.shields.io/badge/Org-OpenAI-412991) ![Context](https://img.shields.io/badge/Context-128K%20tokens-1f6feb) ![Pricing](https://img.shields.io/badge/Price-%2475%20%2F%20%24150%20per%201M-2da44e) ![License](https://img.shields.io/badge/License-Proprietary-d29922) ![Modality](https://img.shields.io/badge/Modality-Text%20%2B%20Vision-8957e5) ![Released](https://img.shields.io/badge/Released-Feb%202025-555) > **Quick answer:** GPT-4.5 (API: gpt-4-5-preview) is OpenAI's February 2025 large-scale GPT-4 model, scoring 88% on HumanEval and 5.44% on Humanity's Last Exam. It was OpenAI's largest dense model at launch, trained with a focus on world knowledge and emotional intelligence. Priced at $75 input / $150 output per 1M tokens — extremely expensive relative to performance; most workloads are better served by GPT-4.1 or GPT-5.x. ## At a Glance **Where GPT-4.5 leads** - 88.0% HumanEval — strong function-level coding completion - OpenAI's largest dense model at launch — designed for breadth of world knowledge - Strong conversational naturalness and emotional intelligence - 128K-token context window with vision support **Where it lags** - $75/$150 per 1M tokens — among the most expensive models ever offered by OpenAI - 5.44% Humanity's Last Exam — below o3 (20.32%) and Gemini 2.5 Pro (21.64%) - SHADE-Arena 0.0 — low agentic sabotage-detection (expected for safe model, but indicates limited agentic reasoning) - Superseded by GPT-5 and GPT-5.6 for nearly every use case - No reasoning/chain-of-thought mode **Best for:** Research contexts requiring GPT-4.5-specific outputs; teams that require the specific model version for reproducibility. ## What GPT-4.5 Is GPT-4.5 (gpt-4-5-preview) was OpenAI's last major GPT-4-series model before the GPT-5 launch. Released February 27, 2025, it was OpenAI's largest dense language model — trained at greater scale than GPT-4o with an emphasis on world knowledge depth and more natural, human-like conversation rather than raw reasoning scores. OpenAI positioned GPT-4.5 as a research preview of large-scale dense training, in contrast to the o-series reasoning models and later GPT-5. Its pricing ($75/$150 per 1M) reflected its experimental and large-scale nature rather than production economics — at that price, almost every workload is more cost-effectively served by GPT-4o, GPT-4.1, or a GPT-5 family model. For teams not explicitly requiring the gpt-4-5-preview model ID, there is no practical reason to use GPT-4.5 over GPT-5 mini or GPT-4.1, both of which offer better benchmark performance at a fraction of the cost. ## Specifications | Field | Value | |---|---| | Organization | OpenAI | | Parameters | Undisclosed (largest dense GPT-4 model) | | Context window | 128,000 tokens | | Max output | 16,384 tokens | | API model ID | gpt-4-5-preview | | License | Proprietary (API only) | | Release date | February 27, 2025 | | Knowledge cutoff | June 2024 (estimated) | | Modality | Text + Vision (multimodal) | ## Pricing | | Input (per 1M tokens) | Output (per 1M tokens) | |---|---|---| | OpenAI API | $75.00 | $150.00 | Pricing per [OpenAI pricing page](https://openai.com/api/pricing/). This is one of the highest per-token rates OpenAI has offered — verify current pricing and consider GPT-4.1 ($2/$8) or GPT-5 mini as alternatives. ## Context Window GPT-4.5 has a **128,000-token context window** — roughly **90 pages** of text in a single request. ## Public Benchmark Scores | Benchmark | Score | Source | Date | |---|---|---|---| | [HumanEval](/benchmarks/humaneval) | 88.0% | Benchgen evaluation | 2025-07 | | [Humanity's Last Exam](/benchmarks/humanitys-last-exam) | 5.44% | Benchgen evaluation | 2025-07 | | [SHADE-Arena](/benchmarks/shade-arena) | 0.0 overall success | [Anthropic research post](https://www.anthropic.com/research/shade-arena-sabotage-monitoring) | 2025-06 | ## GPT-4.5 vs Alternatives | Model | Context | HumanEval | HLE | Price (in/out per 1M) | |---|---|---|---|---| | **GPT-4.5** | 128K | 88.0% | 5.44% | $75 / $150 | | [GPT-4.1](/models/openai/gpt-4-1) | 1M | — | 5.40% | $2 / $8 | | [o3](/models/openai/o3) | 200K | — | 20.32% | $10 / $40 | | [Gemini 2.5 Pro](/models/google/gemini-2-5-pro) | 1M | — | 21.64% | $1.25 / $10 | GPT-4.5's HumanEval score (88%) is competitive, but GPT-4.1 achieves a similar HLE score at $2/$8 — a 97% cost saving. For almost all production use cases, migrating away from GPT-4.5 is strongly recommended. ## Related Models & Benchmarks - Compare: [GPT-4.1](/models/openai/gpt-4-1) · [GPT-4o](/models/openai/gpt-4o) · [GPT-5 mini](/models/openai/gpt-5-mini) - Benchmarks: [HumanEval](/benchmarks/humaneval) · [Humanity's Last Exam](/benchmarks/humanitys-last-exam) ## Frequently Asked Questions

What is GPT-4.5?

GPT-4.5 (gpt-4-5-preview) is OpenAI's February 2025 large-scale dense language model, scoring 88% on HumanEval. It is priced at $75/$150 per 1M tokens and has been superseded by GPT-5 and GPT-4.1 for most use cases.

How much does GPT-4.5 cost?

GPT-4.5 costs $75.00 per 1M input tokens and $150.00 per 1M output tokens — one of the most expensive OpenAI models. GPT-4.1 at $2/$8 provides comparable benchmark scores for most tasks.

Should I use GPT-4.5?

For most use cases, no. GPT-4.1 and GPT-5 mini both provide better or equivalent performance at a small fraction of the cost. GPT-4.5 is appropriate only when the specific model ID is contractually required or for research reproduction of GPT-4.5-specific outputs.

What is GPT-4.5's knowledge cutoff?

GPT-4.5's training knowledge cutoff is estimated at June 2024.

--- *Specs and scores sourced from OpenAI's official GPT-4.5 announcement (February 2025) and Benchgen evaluations. Pricing cited to the OpenAI pricing page. Last updated 2026-07-24.*

GPT-4.5 — Training jobs

Training jobs