OPENAI unifies its model range with GPT-5, accessible to all Chatgpt users. The new model promises increased code and reasoning performance.
Openai promised to unify its AI range, it’s done. The American scale-up unveils this Thursday August 7 GPT-5, its best model to date, for writing tasks, code, or even reasoning. The new family also promises less hallucinations and better precision. The model is available in Chatgpt for all users and via the API for developers.
A new model philosophy
GPT-5 breaks with the fragmented OpenAi approach which hitherto offered different models depending on the use case. No more choice between GPT-4O for speed, O3 for complex reasoning or specialized models according to tasks. The new system incorporates all of these capacities into a unified experience that automatically selects the best approach according to the context of the user query.
A new philosophy of the “unique model” which is accompanied by an evolution in the management of “sensitive” content. Rather than categorically refusing certain requests or responding without nuance, GPT-5 will now adopt a graduated approach called “sure generations”. The model provides the most useful response as possible within the established security limits, clearly explaining the reasons for its possible restrictions. Clearly, the model should never refuse to respond to your requests, the guardrails being directly integrated into the model.
GPT-5 declined in four versions
If GPT-5 adopts a unified approach, Openai offers four different versions of the model, each being adapted to the needs of latency and cost. The main model, GPT-5, Includes the advanced reasoning capacities of the O3 series, the multimodal features of GPT-4O and new agent skills for the execution of multi-stage tasks. GPT-5 Mini And GPT-5 NANO offer lighter alternatives for applications requiring reduced response times.
Finally, GPT-5 PRO Positioned opposite with extensive reasoning, offering more reliable and detailed responses thanks to an elongated processing time, on the model of the O3 series capacities. The company also introduces for developers a control of “the reasoning effort”, making it possible to switch between rapid responses and in -depth analyzes as required. The context window covers 256,000 tokens, compared to 200,000 for GPT-4O. However, we remain well below the million tokens of the Gemini models.
OPENAI presents GPT-5 as its “best code model” and a “real development co-pilot” capable of taking charge of complex implementations on large code bases. GPT-5 is establishing new records on Swe-Bench reference benchmarks (agent programming), Swe-Lancer (evaluation of the precision of the code generated) and helping Polyglot (Multi-Language of the Model). The model manages to further use the tools at its disposal. It could thus become a model of choice for semi-autonomous agent development. The BBVA Spanish Bank teams, which were able to test the preview model, report a double speed compared to the previous models on the code.
GPT-5 … cheaper than GPT-4O
Who says new model also says efficiency gain. Thanks to better optimization, Openai offers GPT-5 at a more affordable price than GPT-4O while the model is more “intelligent”. GPT-5 appears at 1.25 dollars per million tokens as a starter against 2.50 dollars for GPT-4O and 10 dollars out (like GPT-4O). Prices for the PRO, Mini and Nano versions are not yet communicated.
Finally, in Chatgpt, Openai democratizes access to GPT-5 by immediately opening it to all Chatgpt users, including free accounts. Free users will be able to benefit from direct access to a reasoning model for the first time. Enough to feed the growth of OpenAi which claims nearly 700 million active users this week, consolidating its dominant position in the generative AI sector.
