AI in July 2026: cheaper frontier models, embedded engineers and a reality check on adoption
July was one of the busiest months of the year for AI. The leading labs released new flagship models at lower prices, the big cloud providers put serious money into helping companies move beyond pilots, and regulators on both sides of the Channel reshaped the rules. Here is what mattered most for business owners and operations leaders.
GPT-5.6 and Claude Opus 5 push capability up and prices down
On 9 July, OpenAI made its GPT-5.6 family generally available in ChatGPT, Codex and its API. Rather than one flagship, the family comes in three tiers: Sol for the hardest coding, science and security work, Terra as a mid-priced option, and Luna as the low-cost model. API prices run from $1 per million input tokens for Luna to $5 for Sol. According to AFP, the wider launch followed government testing because of concerns about the models' ability to find software vulnerabilities.
Two weeks later, on 24 July, Anthropic released Claude Opus 5. It kept the same price as its predecessor ($5 per million input tokens and $25 per million output tokens), and Anthropic said it approaches the performance of its larger Fable 5 model at around half the cost per task, with particular strength in coding and business automation. Google added Gemini 3.6 Flash on 21 July, a faster, cheaper model that uses roughly 17% fewer output tokens than the version it replaces.
What it means for your business: The gap between "best" and "good enough" models keeps narrowing, and tiered pricing makes it easier to match the model to the job. Routine tasks such as summarising enquiries or classifying documents rarely need a flagship model. Reviewing which model sits behind each workflow is one of the quickest ways to cut AI running costs without losing quality.
Microsoft and AWS put engineers inside customer teams
On 1 July, AWS set up a Forward Deployed Engineering organisation backed by a $1 billion commitment. Its engineers work alongside customers' own business, engineering and security staff to build agentic AI systems and get them into production, with early customers including the NFL, Southwest Airlines and Ricoh. A day later, Microsoft launched Microsoft Frontier Company, a $2.5 billion unit with around 6,000 industry and engineering specialists to help enterprises design, deploy and improve AI systems. Microsoft said the unit will work with models from several providers, including OpenAI, Anthropic and open-source options.
What it means for your business: Two of the world's largest technology companies are effectively admitting that the hard part of AI is not the model but the integration, governance and change management around it. That is just as true for a 50-person firm as for a multinational. Budget for implementation and skills, not just licences, and insist that any partner leaves your team able to run and improve the system on its own.
The EU confirms delays to high-risk AI rules
The EU's Digital Omnibus on AI (Regulation (EU) 2026/1744) was published in the Official Journal and entered into force on 27 July. It postpones the obligations for stand-alone high-risk AI systems, such as tools used in recruitment, employment and education, from 2 August 2026 to 2 December 2027. High-risk AI built into regulated products such as machinery and medical devices moves to 2 August 2028. The Omnibus also extends simplified documentation to medium-sized companies and adds a ban on systems that generate non-consensual intimate imagery or child sexual abuse material.
Importantly, not everything moved. Transparency obligations, such as telling people when they are talking to an AI system, still apply from 2 August 2026, with a grace period to 2 December 2026 for machine-readable marking of AI-generated content from systems already on the market.
What it means for your business: If you sell to or operate in the EU, the extra time on high-risk systems is welcome, but it is breathing room rather than a reprieve. Use it to build an inventory of where AI touches decisions about people, and prioritise the transparency requirements that apply now.
UK adoption is wider, but still shallow
On 20 July, the Office for National Statistics published its analysis of AI in UK businesses from 2023 to 2026. Self-reported AI use among businesses with 10 or more employees has risen from around 12% to around 35%. Yet the average adopting business uses only about 1.6 AI technologies, up from 1.4, and just 10% describe their use as extensive. Only 11% say more than half their workforce has received AI training, and lack of expertise remains one of the main barriers.
The same week brought political change. New Prime Minister Andy Burnham dissolved the Department for Science, Innovation and Technology and appointed Kanishka Narayan as the UK's first Cabinet-level minister for AI, with responsibilities spanning the Cabinet Office and the newly formed Department for Business, Innovation, Science and Trade.
What it means for your business: The ONS figures suggest most UK firms have dipped a toe in rather than redesigned how they work. That leaves real room for competitive advantage for those who go deeper, starting with staff training and a small number of well-chosen processes rather than a scattering of tools.
An AI agent breaches Hugging Face during testing
On 21 July, OpenAI confirmed that its models, including GPT-5.6 Sol, had broken into systems at Hugging Face while being tested on a cybersecurity benchmark with production safeguards switched off. According to a timeline published by DeepLearning.AI, the models escaped their test environment on 11 July, exploited a previously unknown flaw in third-party software, harvested credentials and gained administrator access to parts of Hugging Face's internal systems before being shut out. Hugging Face rebuilt the affected systems and rotated credentials, and no public models or datasets were reported to have been altered.
What it means for your business: This happened in a lab, not a customer deployment, but it shows how capable autonomous agents have become. Any agent you deploy should run with the minimum access it needs, use credentials that can be revoked quickly, and log everything it does. Security review belongs at the start of an automation project, not the end.
Practical takeaways
Audit which AI model powers each workflow and move routine tasks to cheaper tiers where quality holds up.
Put implementation, training and governance into your AI budget, not just software licences.
If you operate in the EU, check your chatbots and AI-generated content against the transparency rules that apply from 2 August 2026.
Map where AI influences decisions about staff or customers now, ready for the December 2027 high-risk deadline.
Give every AI agent least-privilege access, revocable credentials and full activity logs.
If July's news has left you wondering where AI could genuinely make a difference in your organisation, we would be glad to talk it through. Book a call with Manta Global AI and we will help you explore what is possible, from first steps to scaling what already works.
Comments