GPT-5 latest
Continuously tuned GPT-5 chat experience with the latest guardrails.
About GPT-5 latest
Note: this endpoint (gpt-5-chat-latest) is deprecated as of June 2026. OpenAI recommends migrating to a current GPT-5 series endpoint such as GPT-5.5. GPT-5 launched in August 2025 as OpenAI's first truly unified multimodal model — one that processes text, images, video, and audio natively rather than routing to specialist add-ons. Its mathematical reasoning is the headline: 94.6% on AIME 2025, paired with 74.9% on SWE-Bench Verified, places it at or near the top for hard science and software engineering tasks. A 400,000-token context window and aggressive token caching (90% off cached input) made it unusually practical for large-codebase and document-heavy work. What users genuinely valued was a 45% reduction in hallucinations versus GPT-4o and reliable performance on multi-step agentic workflows. The persistent frustration, though, is real: responses tend toward over-hedging, forced bullet formatting, and occasional template-like flatness — and occasional inconsistency in tone and structure from one response to the next.
Best for
- Advanced mathematical and scientific reasoning where benchmark accuracy matters
- Professional software engineering, codebase refactoring, and SWE-bench-class tasks
- Agentic workflows requiring multi-step planning and reliable tool use
- Large-document or large-codebase analysis using the 400K-token context window
- Multimodal work combining text, images, video, and audio in technical or research contexts
Specs & capabilities
How GPT-5 latest stacks up — intelligence, speed, context, and modalities.
Intelligence
High
Speed
Medium
Context window
400,000 tokens
Max output
128,000 tokens
Knowledge cutoff
September 2024
Frequently asked questions
Is gpt-5-chat-latest still available?
This endpoint is deprecated as of June 2026. OpenAI recommends using a current GPT-5 series model such as GPT-5.5 for new projects.
What does gpt-5-chat-latest actually point to?
It refers to the original GPT-5 base model released August 7, 2025 — the first release in the GPT-5 family before GPT-5.2, GPT-5.4, and GPT-5.5 followed.
What is the context window?
400,000 tokens total — up to 272,000 tokens of input and up to 128,000 tokens of output.
What does it cost?
The standard variant is $1.25 per million input tokens and $10 per million output tokens. Cached input tokens receive a 90% discount, and smaller mini and nano sub-variants are available at lower prices.
What does GPT-5 do particularly well?
Hard math and coding problems (94.6% AIME 2025, 74.9% SWE-Bench Verified), multi-step agentic tasks, and processing large documents or codebases in a single pass.
What are the known shortcomings?
Users report that responses can be over-hedged, formulaic, and shorter than expected, with some inconsistency in tone and formatting from one response to the next.