OpenAI has released GPT-5 to all ChatGPT paid subscribers, introducing real-time web access and a significantly upgraded voice mode as standard features across Plus and Team plans. The rollout began Wednesday and is expected to reach all eligible accounts within 48 hours, marking the most significant update to the platform since GPT-4o launched in 2024.
GPT-5 replaces the previous Search plugin with a fully integrated browsing capability that retrieves live information without any additional configuration. OpenAI said the model processes web results differently from its predecessor — rather than returning raw excerpts, GPT-5 synthesises multiple sources into a coherent answer while citing its references inline. The company described the improvement as reducing the need for users to cross-reference responses with separate searches.
In internal evaluations, OpenAI said GPT-5 handles multi-step reasoning and software engineering tasks with fewer errors than GPT-4o, and that reasoning latency has been reduced by approximately 30 percent. The gains are attributed to architectural changes made during training rather than increased model size, which OpenAI said has remained comparable to its previous flagship.
Voice 2: emotional tone detection and natural interruptions
The new voice mode, called Voice 2, is the most substantial upgrade to the audio interface since the feature launched. Voice 2 introduces real-time emotional tone detection, allowing the model to recognise stress, frustration, or enthusiasm in the user’s speech and adjust its response style accordingly. OpenAI demonstrated the feature in a live session handling mid-sentence interruptions and abrupt topic changes without losing context — behaviour that previous voice implementations handled poorly.
Voice 2 is currently available in English only. OpenAI said additional language support — including Spanish, French, Mandarin, and German — is planned for the fourth quarter of 2026. The company has not confirmed whether the emotional detection feature will be extended to all language variants or remain English-only initially.
Free tier access and API implications
Free tier users will gain access to a limited version of GPT-5 with daily usage caps. OpenAI did not specify the cap thresholds, but said they are designed to allow meaningful evaluation of the model’s capabilities without full production access. The company confirmed that GPT-4o will remain available to free users who prefer it or reach their GPT-5 allocation for the day.
For API customers, GPT-4o remains the default model. OpenAI has not announced a timeline for GPT-5’s general API availability, though enterprise customers under direct agreements are expected to gain access sooner. The API pricing structure for GPT-5 has not been officially confirmed.
Competitive context
The release arrives as competition in the large language model market has intensified significantly over the past 12 months. Anthropic’s Claude 4 Opus, available through API and consumer subscriptions, has been widely benchmarked as competitive with GPT-4o on coding and reasoning tasks. Google’s Gemini 2.0 Ultra is available through Workspace and enterprise channels, and Meta’s open-source Llama 4 series has expanded the range of capable models available without usage fees.
OpenAI said GPT-5 outperforms all publicly available models on its internal benchmark suite, though it acknowledged that external independent evaluations will determine how those claims hold up against real-world usage patterns. Third-party benchmark results from organisations including HELM and BIG-bench are expected to be published within weeks of the model’s general availability.
