IGI Logo

GPT-5 unleashed: Presents risks and opportunities

OpenAI’s GPT-5 might not present a step-jump in capabilities such as GPT-4, but it claims to offer a meaningful improvement in reasoning and coding tasks, among others. Hallucination rates have dropped significantly, a positive for gen AI and agentic AI adoption. GPT-5 will increase the adoption of gen AI in software development, exposing IT services firms to revenue deflation risks, but it is also a step forward in opening up new opportunities from (1) cloud and data foundation, (2) legacy modernization and (3) AI for business use cases. 

GPT-5 lands finally but not a step jump compared to recent predecessors

OpenAI released GPT-5 series of models after much anticipation. Models have 400k context length, with 128k max output tokens. The models incorporate elements of predecessors such as reasoning but do not seem to possess capabilities that could be categorized as a step-jump. The models carry several improvements that aim to provide a better user experience. GPT-5 claims significant advances in reducing hallucinations, improving instruction following, and minimizing sycophancy, while leveling up performance in three of ChatGPT’s most common uses: writing, coding and health.

Presents improvements in coding tasks

GPT-5 is OpenAI’s strongest coding model to date. OpenAI claims particular improvements in complex front?end generation and debugging larger repositories and notes that it is state-of-the-art (SOTA) across key coding benchmarks, scoring 74.9% on SWE-bench Verified and 88% on Aider polyglot. According to OpenAI, GPT?5 was trained to be a true coding collaborator. It excels at producing high-quality code and handling tasks such as fixing bugs, editing code and answering questions about complex codebases. The model is steerable and collaborative—it can follow very detailed instructions with high accuracy and can provide upfront explanations of its actions before and between tool calls. GPT?5 also excels at long-running agentic tasks—achieving SOTA results on τ2-bench telecom (96.7%), a tool-calling benchmark released just two months ago. GPT?5’s improved tool intelligence lets it reliably chain together dozens of tool calls—both in sequence and in parallel—without losing its way, making it far better at executing complex, real-world tasks end to end. It also follows tool instructions more precisely, is better at handling tool errors, and excels at long-context content retrieval. Prominent code editors such as Cursor and Windsurf have rated GPT-5 highly relative to others (Exhibit 2).

Reduction in hallucination rates and lower pricing are positives for AI adoption

Hallucination rates have reduced significantly (Exhibit 1) compared to predecessors and thus will help quell worries of hallucinations increasing with an increase in model capabilities. Lower hallucinations help increase the reliability of generative AI tools and are a positive for adoption across use cases, especially in agentic AI. Improvements in other categories such as intelligence also help in this regard. Pricing per token is lower compared to frontier models from other key US-based companies such as Anthropic and Google (Exhibit 3).

Raises risks and opportunities for IT services—expect net headwinds in the near term

We believe that GPT-5 is another step toward increasing generative AI adoption among enterprises. Given (1) the focus of AI labs to increase coding capabilities, (2) the willingness of enterprises to adopt generative AI in software development, (3) the increasing developer usage of generative AI tools and (4) the higher focus of enterprises to get productivity from AI adoption, we believe that the adoption of generative AI in software development will increase and present revenue deflation risks for IT services providers. The same applies for customer service BPO as well.

At the same time, reduction in hallucinations, lower pricing, better reasoning and intelligence, among other improvements, also open the door for more and faster adoption of AI for business use cases. These can present opportunities for Indian IT. They can help enterprises set up cloud and data foundation required to reap the full value of generative AI. Effort, costs and risks around legacy modernization can be reduced, leading to accelerated modernizations. IT services firms can be partners providing end-to-end services for viable AI-enabled use cases for business. Higher velocity and lower costs of software development can catalyse higher custom application builds.

We believe that new opportunities from generative AI adoption will offset revenue deflation in existing volumes over time. However, we expect a lag in the pickup of new opportunities and for savings in software development to be deployed into these new opportunities, leading to a period of net headwinds. We believe gen AI can impact revenue growth for Indian IT by 2-3% for a period of 2-3 years on a net basis. Our current revenue growth assumptions build in some portion of this impact already.

Can boost custom app development, providing an alternative to SaaS; may require push from Indian IT

Increased velocity and lower costs of software development bode well for custom application development. Custom apps help enterprises utilize contextual knowledge and enable differentiation as opposed to SaaS, which promotes standardization. Enterprises will also benefit from lower security/data issues, lower vendor lock-in risks and higher customization possibilities from reduced reliance on SaaS. Boosting custom app development might require evangelism on the part of Indian IT vendors while not offending SaaS partners.

We expect IT services to continue to be relevant even in a future with high gen AI adoption

At a fundamental level, service provides one-on-one customizable solutions compared to software/AI, which provides a one-to-many solution. One can argue that a sufficiently capable AI can itself provide customized solutions, provided it is well fed with enterprise data and context and can take on the role played by services. We do not expect everything in the enterprise to be measurable and hence possible to be given as an input to AI or for what is measured to be truly representative of the underlying due to the potential of it being gamed. There will continue to be scope for services, perhaps it can get redefined over a period of time.

The value that gets captured by services compared to other technology elements (software/cloud infra/AI model providers, etc.) will be a key point of debate, though. We do note that improvements in generative AI capabilities by an AI lab tends to get quickly replicated by others despite the high costs involved and scarcity of quality AI talent. There is intense competition among at least 5-6 players globally (mainly in the US and China), ensuring there is no lasting moat. Improvements in hardware will also likely lead to a reduction in training and inference costs. Such an environment is ideal for services, since it will ensure that AI models do not end up capturing disproportionately high value (assuming perfect competition). 

Above views are of the author and not of the website kindly read disclaimer