OpenAI

OpenAI's newest flagship general-purpose model (April 2026). Excels at complex reasoning, coding, and creative tasks. Features a 1M+ token context window. Cached input drops 90% (to $0.50/MTok). Data residency endpoints incur a 10% price uplift.

Maximum capability variant of GPT-5.5 for research-grade and highest-stakes tasks. No cached-input discount available. Data residency endpoints incur a 10% price uplift.

Primary advanced reasoning model. Replaced o1 ($15/MTok) with an 87% price cut and improved performance. Excels at coding, math, science, and visual perception. Internal reasoning tokens are billed at output rates, which can multiply actual costs by 3-10x depending on task complexity.

Lightweight reasoning model. Positioned between o4-mini and o3 for cost/capability trade-offs. Context window expanded to 200K tokens.

Maximum reasoning capability in OpenAI's lineup. For workloads where correctness is paramount over cost. No cached-input or batch discount available.

Most cost-effective option for 1M-token context workloads. Best choice for long-document processing when cost matters. Reliable production model.

Recommended production workhorse. Strong general-purpose frontier model at half the price of GPT-5.5 with 1M context window. Cached input at $0.25/MTok (90% discount). Data residency endpoints incur a 10% price uplift.

Pro variant of GPT-5.4 for maximum capability within the 5.4 family. No cached-input discount. Data residency endpoints incur a 10% price uplift.

Legacy flagship multimodal model. Still available for legacy integrations and multimodal (text + image + audio) use cases. Supports the Realtime API for low-latency voice. GPT-5 ($1.25/MTok) is now generally more capable at a lower price.

Previous flagship model. Still available and capable; GPT-5.4 has largely replaced it for production use cases. Cheaper than GPT-4o on input ($1.25 vs $2.50) with generally higher capability.

Previous flagship image generation model, now superseded by GPT Image 1.5. Still available and capable. Slightly more expensive than 1.5 at High quality tier. Pricing also available per-token (text + image tokens billed separately) via the Responses API. Génération image via RodiumAI POST /v1/images/generations (OpenAI upstream).

Current flagship image generation model. Best quality in OpenAI lineup (LM Arena Elo 1264). Supports Low/Medium/High quality tiers and text-to-image + image editing. Includes C2PA metadata in generated images. Also accessible via alias chatgpt-image-latest. Génération image via RodiumAI POST /v1/images/generations (OpenAI upstream).

Best value for long-context tasks at low cost. 1M context window at only $0.40/MTok input.

Cheapest model in OpenAI's lineup. $0.10/MTok input is competitive with the cheapest models on the market. Ideal for routing, extraction, and lightweight classification. Remains competitive even after GPT-5.5 launch.

Legacy budget option with wide ecosystem support and 128K context. GPT-4.1 Nano ($0.10/MTok) or GPT-5.4 Nano ($0.20/MTok) are the recommended modern replacements. Web search tool charges a fixed block of 8,000 input tokens per call for this model.

Best price/performance for production apps. One of the most competitive mid-tier models on the market — 4x cheaper than Claude Sonnet 4.6 on input. Outperforms GPT-4 Turbo at 90%+ lower cost. Data residency endpoints incur a 10% price uplift.

Ultra-budget option for routing, extraction, and high-volume backend tasks. 12x cheaper on input than GPT-5.4. Data residency endpoints incur a 10% price uplift.

Balanced cost and capability from the GPT-5 family. Good for chatbots and mid-tier production workloads.

Budget-friendly image generation. 50-70% cheaper than GPT Image 1.5. Cheapest entry point for API image generation at /bin/sh.005/image (low quality). Recommended for high-volume projects, social media, and testing. Génération image via RodiumAI POST /v1/images/generations (OpenAI upstream).

Most cost-effective reasoning model. Standout value for math, logic, and code at scale. Internal reasoning tokens billed at output rates. Frequently combined with GPT-5.4 Mini in multi-model agent architectures.