GPT-5 fits teams already running OpenAI workflows that need frontier coding and math performance, scoring 97.4% on HumanEval code generation. It replaced GPT-4o as OpenAI's default flagship until GPT-5.5 took over in 2026. Teams wanting OpenAI's current best performance should choose GPT-5.5 instead; GPT-5 stays usable only until its scheduled API removal.
GPT-5 is OpenAI's Mixture-of-Experts flagship language model, scoring 94.6% on the AIME math competition and built for coding, reasoning, and multimodal tasks. It processes text, images, audio, video, and PDFs in a single API call. GPT-5 has since been succeeded by GPT-5.5 as OpenAI's current flagship model.
Where it sits
- $1.72/M$ per 1M tokensBlended price (3:1)Lower is better#32 / 62peer median $1.72/Mvendor price, checked by HokAI
- 75 tok/stokens/sOutput speedHigher is better#23 / 37peer median 90 tok/scited: Artificial Analysis
- 74.9%% solvedSWE-bench VerifiedHigher is better#19 / 27peer median 78%per source, see benchmark scores
- 88.4%% correctGPQA DiamondHigher is better#22 / 44peer median 88.1%per source, see benchmark scores
Priced around the middle of the 62 GA models with a published price (rank 32), in the bottom third on SWE-bench Verified (rank 19 of 27), and one of 22 that document a zero-data-retention option. Ranked against GA models; this record is not GA.
Ranks are against GA models on HokAI that publish the same figure; ties share a rank.
Provider: OpenAI · Family: GPT-5
Context window: 272,000 tokens · Max output: 128,000
Input modalities: text, image, audio, video, pdf, tool-calls · Output: text, audio, tool-calls
About GPT-5
GPT-5 is OpenAI's fifth-generation foundation model, released publicly on August 7, 2025 as the successor to GPT-4o. It is OpenAI's first flagship built on a sparse Mixture-of-Experts architecture rather than the dense transformer design used across the GPT-4 lineage. OpenAI has not disclosed the total parameter count; industry estimates place it at 2 to 5 trillion parameters, with only a fraction active per forward pass through expert routing.
OpenAI published the GPT-5 System Card on August 13, 2025, evaluating the model against biological, chemical, nuclear, and radiological uplift scenarios under its Preparedness Framework before deployment approval. The system card also documents alignment work that reduces sycophancy through additional post-training beyond standard RLHF.
GPT-5 was followed by a rapid succession of updates: GPT-5.1 in October 2025, GPT-5.2 in December 2025 with a 400K-token context window, GPT-5.4 in early 2026 with native computer use, and GPT-5.5 in April 2026, which became OpenAI's current flagship. The original snapshot (gpt-5-2025-08-07) was deprecated June 11, 2026, with API removal scheduled for December 11, 2026; teams still relying on it should migrate to GPT-5.5 before that cutoff.
Pricing
$0.625 per 1M input tokens, $5.00 per 1M output. Batch API at 50% off: $0.3125 input and $2.50 output. Flex processing also at 50% off with variable wait times.
What a real job costs
| Job | Input | Output | Total |
|---|---|---|---|
| Summarise a 20-page PDF | $0.019 | $0.0050 | $0.024 |
| Support reply | $0.0013 | $0.0015 | $0.0027 |
| One coding agent run | $0.125 | $0.100 | $0.225 |
Budgets: 20-page PDF = 30k in / 1k out · Support reply = 2k in / 300 out · Coding agent run = 200k in / 20k out. Computed from the vendor's per-token prices at render time; cached-input discounts are not applied.
Key Features
- Sparse MoE Architecture: Routes each input to specialized expert subnetworks instead of activating all parameters, which is how OpenAI holds coding and reasoning accuracy at a comparatively low per-token API price.
- 272K-Token Context Window: Accepts up to 272K input tokens in the API, doubling GPT-4o's context limit, with a 128K-token cap on output per request.
- Math and Reasoning Benchmarks: Scored 94.6% on AIME 2025 math competition problems and 88.4% on GPQA Diamond graduate-level science reasoning with extended thinking enabled.
- Native Multimodal Input: Processes text, images, audio, video frames, and PDFs within a single API call through an integrated vision encoder rather than a separate preprocessing pipeline.
- Parallel Tool Calls: Executes multiple function calls in one response turn with structured JSON output; code execution is available only in the ChatGPT interface's built-in interpreter, not the raw API.
Pros
- Scored 74.9% SWE-bench Verified at launch, among the strongest real-world coding results OpenAI had published at the time.
- Sparse MoE routing kept frontier-tier accuracy available at a comparatively low per-token API price without a separate lower-cost model.
- Extended thinking mode lifts GPQA Diamond graduate-level science reasoning to 88.4%, one of the largest reasoning-mode gains OpenAI has published.
Cons
- Superseded by GPT-5.5 (April 2026), which scores 88.7% SWE-bench Verified and adds a 1M-token context window.
- No self-hosting option; proprietary closed weights require API dependency, ruling out air-gapped and on-device deployments.
- The original GPT-5 snapshot is deprecated (June 11, 2026) and loses full API access on December 11, 2026, so new integrations should target GPT-5.5 instead.
Benchmarks
- MMLU: 91.4% vendor-reported · 07 Aug 2025 — General-knowledge exam across 57 subjects, % correct.
- AIME 2025: 94.6% vendor-reported · 07 Aug 2025 — Competition-level maths problems from the 2025 exam, % solved.
- HumanEval: 97.4% vendor-reported · 07 Aug 2025 — Small programs that must pass hidden tests, % passing.
- GPQA Diamond: 88.4% vendor-reported · 07 Aug 2025 — PhD-level science questions that are hard to search for, % correct.
- Aider Polyglot: 88% independent · 15 Aug 2025 — Code edits across several programming languages, % correct.
- SWE-bench Verified: 74.9% vendor-reported · 07 Aug 2025 — Real GitHub issues fixed end to end, % solved.
- AA Intelligence Index: 45 cited: Artificial Analysis — Composite of 10 evaluations run by Artificial Analysis, 0 to 100.
- AA blended price: $2.81/M cited: Artificial Analysis — Price per 1M tokens at a 3:1 input to output blend, as listed by Artificial Analysis.
- Output speed: 75 tok/s cited: Artificial Analysis — Median tokens written per second as measured by Artificial Analysis.
A benchmark is an exam, not the job. Scores transfer unevenly between tasks, so weigh the one closest to your workload and read every figure with its source.
Frequently Asked Questions
How much does GPT-5 cost in 2026?
GPT-5 costs $0.625 per million input tokens and $5.00 per million output tokens on the standard pay-as-you-go tier. The Batch API halves both rates to $0.3125 input and $2.50 output, with results delivered asynchronously within 24 hours. Flex processing offers the same 50% discount at variable processing speed.
Is GPT-5 free to use?
GPT-5 has no free API tier; every request is billed per token at the standard rate. Inside ChatGPT, free-tier users get an 8K-token context window on GPT-5 responses, and Plus subscribers get 32K, well below the limit available to API users or Pro subscribers.
What are the best alternatives to GPT-5?
The most direct alternative is GPT-5.5, OpenAI's current flagship, which scores higher on coding benchmarks and adds a larger context window. Outside OpenAI, Claude Opus 4.8 is a comparable frontier option for long-horizon coding and reasoning work. Teams that need self-hosting or offline deployment should look at gpt-oss-120b instead, OpenAI's separate open-weight model.
What separates GPT-5 from Claude Opus 4.8?
Claude Opus 4.8 scores 88.6% on SWE-bench Verified with a 1M-token context window, a step up from GPT-5's coding and context performance at launch. Anthropic charges $5 per million input tokens and $25 per million output tokens, a different pricing structure than GPT-5's flat per-token API rate. Pick Claude Opus 4.8 for the strongest available reasoning depth today; pick GPT-5 only if you already have OpenAI-specific integrations in place.
What does it take to start using GPT-5?
Generate an API key at platform.openai.com and authenticate requests with a bearer token; Azure customers can instead use Azure Active Directory or an API key through Azure AI Services. The Python, TypeScript, JavaScript, Go, Java, and .NET SDKs all support the model out of the box. Since GPT-5 has been superseded, new projects should default to GPT-5.5 unless they specifically need to match GPT-5's original benchmarked behavior.
Top Alternatives
- GPT-5.5: Pick GPT-5.5 if you need OpenAI's current-generation performance; it scores 88.7% SWE-bench Verified with a 1M-token context window versus GPT-5's original spec.
- Claude Opus 4.8: Pick Claude Opus 4.8 if raw reasoning depth matters most; GPT-5 remains viable only for teams already standardized on OpenAI's API and tooling.
HokAI guides covering GPT-5
- ChatGPT Can Still Shop For You. It Just Can't Check You Out Anymore.: ChatGPT's shopping research still works well, but Instant Checkout died in March 2026. Here is how ChatGPT, Perplexity, and Gemini compare for buying now.
- OpenAI Killed Sora in March. Five Months Later, Here's Who Actually Won.: Sora's app is gone, its API dies September 24, and Disney never named a replacement. Here's what actually happened since OpenAI's March shutdown announcement.