LLM Models

DISCONTINUED. OpenAI's cost-efficient reasoning model optimized for STEM tasks. Uses same high-compute reinforcement learning pipeline as o1-preview but with smaller, faster model. Achieves 70% on AIME math competition (competitive with o1's 74.4%), placing it in top 500 US high school students. Reaches 1650 Elo on Codeforces (86th percentile of programmers), nearly matching o1's 1673 Elo. Generates answers 3-5x faster than o1-preview while being 80% cheaper. Excels at STEM reasoning but performs worse on non-STEM factual knowledge. Recommended upgrade to o3-mini available. OpenAI removed it from the API on October 27, 2025, having announced the deprecation on April 28, 2025.

OpenAI's most cost-efficient small model delivering intelligent performance at breakthrough pricing. Despite compact size, surpasses GPT-3.5 Turbo across academic benchmarks including 82% on MMLU (textual intelligence), 87% on MGSM (math reasoning), and 87.2% on HumanEval (coding). Features 128K context window and supports both text and vision inputs. Priced at 15 cents per 1M input tokens (60% cheaper than GPT-3.5 Turbo), making it ideal for high-volume applications requiring reliable intelligence without flagship costs.

DISCONTINUED. OpenAI's advanced text-to-image generation model with exceptional prompt adherence. Represents a major leap from DALL-E 2 in understanding complex and lengthy text prompts. Integrated with ChatGPT to automatically generate detailed, tailored prompts from simple user ideas. Produces high-quality images with improved handling of historically difficult elements like hands and text within images. Available in Standard and HD quality modes. OpenAI has deprecated DALL-E 3 and removed it from the API.

OpenAI's previous-generation flagship with vision capabilities and large context window. Represents a significant upgrade from base GPT-4 with 3x cheaper input pricing and 2x cheaper output pricing. Features 128K context window (equivalent to 300+ pages) and can generate up to 4,096 tokens per response. Strong performance on reasoning, math, and coding tasks with native vision support for analyzing images.

OpenAI's cheapest and fastest model in the GPT-4.1 family (April 2025), built for high-volume classification, extraction, and autocomplete where sophisticated reasoning isn't needed. Retains the 1M-token context window at the lowest price in the lineup.

The mid-tier model of OpenAI's GPT-4.1 family (April 2025), matching or beating GPT-4o on many benchmarks at roughly 5x lower cost, with the same 1M-token context window. A strong default for latency- and cost-sensitive production workloads.

OpenAI's April 2025 workhorse API model, a major step up over GPT-4o on coding, instruction following, and long-context comprehension, with a 1M-token context window. GPT-4.1 is ~26% cheaper than GPT-4o for median queries and is offered alongside smaller mini and nano tiers.

OpenAI's versatile flagship model with multimodal capabilities. Accepts text and image inputs, produces text outputs. The 'o' stands for 'omni', reflecting its ability to handle multiple input types seamlessly. Features a massive 128K context window and strong performance across reasoning, coding, and creative tasks. Best model for most use cases requiring balanced performance and cost.

DISCONTINUED. OpenAI's first reasoning-focused model that uses extended chain-of-thought processing to tackle complex problems. Spends more time thinking before responding, breaking down problems systematically. Achieves exceptional performance on PhD-level science questions (GPQA) and competitive programming (Codeforces 89th percentile). Best for tasks requiring deep reasoning: mathematics, scientific analysis, coding challenges, and multi-step problem decomposition. OpenAI removed it from the API on July 28, 2025, having announced the deprecation on April 28, 2025.

OpenAI's flagship model optimized for coding and agentic workflows with adaptive reasoning. Two variants (Instant and Thinking) balance speed and thoroughness. Instant tier features breakthrough adaptive reasoning that automatically decides when to think deeply versus respond quickly. Thinking tier provides persistent reasoning for complex multi-step problems. Strong improvements on math and coding benchmarks including AIME 2025 and Codeforces.

OpenAI's late-2025 flagship (December 2025) with a massive 400K context window and enhanced reasoning capabilities, now superseded by the GPT-5.4 and GPT-5.5 generation (April 2026). Available in three tiers (Instant, Thinking, Pro) for different use cases. First model to exceed 90% on ARC-AGI benchmark, representing a significant leap in general intelligence. Excels at complex reasoning, coding, and agentic workflows with hybrid reasoning that balances speed and depth.

OpenAI's state-of-the-art image generation model, producing high-fidelity images with strong prompt adherence and text rendering. Successor to GPT Image 1, available through the API.