Microsoft’s CoreAI organization has told employees working on AI coding projects to default to OpenAI’s GPT-5.6 Sol in GitHub Copilot, according to an internal memo from Executive Vice President Jay Parikh viewed by CNBC. The instruction is an internal efficiency policy, not a new GitHub Copilot default for customers—and that distinction is the important one for IT teams watching Microsoft’s AI spending and model strategy.
CNBC reports that Parikh told staff that moving more workloads to OpenAI models delivers greater value from Microsoft’s token investment. The wording ties the choice directly to Microsoft’s commercial relationship with OpenAI, rather than presenting GPT-5.6 Sol as a universal technical winner for every coding task. Microsoft has not publicly released the memo, its model-routing criteria, or any measurement showing how much token or developer-time savings the directive is expected to produce.
For developers outside Microsoft, the immediate practical change is none: GPT-5.6 Sol remains a selectable Copilot model where the relevant plan and organization policy permit it. GitHub’s own release documentation says Sol is intended for complex reasoning across large codebases and demanding, long-running agentic work; it calls GPT-5.6 Terra the balanced model for day-to-day interactive and agentic coding, and GPT-5.6 Luna the low-cost option for smaller, faster tasks.
That makes Parikh’s instruction a revealing internal economic decision, not a public declaration that every GitHub Copilot user should select Sol.
The memo arrives after Microsoft and OpenAI rewrote the terms of their partnership earlier in 2026. OpenAI said in April that Microsoft retains a license to OpenAI models and products through 2032, while Microsoft continues receiving revenue-share payments from OpenAI through 2030, subject to a cap. Microsoft also remains a major OpenAI shareholder.
Those facts give Microsoft a reason to treat OpenAI usage differently from usage of Anthropic, Google, open-weight models, or even its own developing model families. A dollar spent on OpenAI inference is not economically identical to a dollar sent to an unrelated provider. Microsoft’s licensing rights let it build OpenAI technology into products such as Copilot, and its continuing financial arrangements mean growing OpenAI usage can have benefits beyond the immediate compute bill.
Microsoft’s 2025 annual filing stated that it holds rights to OpenAI intellectual property—including models and infrastructure—for integration into its products. The revised 2026 arrangement made the license non-exclusive, but did not remove it. Parikh’s reported direction reflects the operational implication: when a workload can be handled by an OpenAI model that Microsoft is already entitled to deploy and from which it can still receive economic value, sending that workload elsewhere may be harder to justify internally.
The company has simultaneously been building its own AI capabilities. Microsoft created the CoreAI – Platform and Tools organization under Parikh in January 2025 to unite developer-facing platforms including GitHub, Visual Studio, Visual Studio Code, Azure AI Foundry, and related engineering work. The existence of that division makes this memo more consequential than a routine employee recommendation. It concerns the groups that shape how many developers encounter AI tooling in practice.
Yet the direction should not be misread as Microsoft abandoning a multi-model approach. GitHub Copilot continues to expose choices from multiple model providers, and Microsoft’s public documentation for Copilot-based development workflows still recommends current Claude Sonnet, Claude Opus, OpenAI Codex, and GPT-5-class models depending on the task. Microsoft’s external posture remains model choice; the reported internal posture is to prefer the partner model when the economics support it.
The most plausible explanation is commercial, not technical. A company writing, reviewing, and maintaining the developer products themselves may value top-tier reasoning quality on internal projects more highly than it values per-task cost. But Parikh’s stated rationale, as reported by CNBC, was explicitly about extracting more value from Microsoft’s token investment in OpenAI. That is a business optimization layered over engineering preferences.
For an enterprise running GitHub Copilot, this is a useful warning against treating an employee memo as deployment guidance. A model that makes sense for Microsoft’s own internal economics can be a poor default for a customer whose Copilot bill is calculated from provider list pricing and whose simpler tasks dominate volume.
GitHub’s current model catalog does not make Sol universally available, either. Sol is offered on Copilot Pro+, Max, Business, and Enterprise plans, while Terra and Luna have broader availability that includes the standard Pro tier. For Business and Enterprise customers, GPT-5.6 access must be enabled by an administrator; GitHub says that policy is off by default. Availability is also rolling out gradually across Copilot surfaces.
So even if a CIO wanted to mirror Microsoft’s reported internal default, many organizations would first need to enable the model, confirm their plan supports it, and decide whether that is justified by their workload profile.
Under that structure, the model picker is no longer merely a quality setting. It is a spending control.
GitHub’s July release notice says GPT-5.6 models are billed at provider list pricing through Usage Based Billing. Microsoft has not publicly said whether its internal teams pay those same rates through internal chargeback systems, receive preferential accounting treatment, or have a separate arrangement for OpenAI capacity. That missing detail is central to the memo’s logic: the company’s claimed “greater value” may arise from partnership economics that a normal Copilot customer cannot reproduce.
Administrators should therefore resist a blunt policy of forcing every developer onto the biggest available model. Sol may be appropriate for repository-wide refactors, architecture reviews, complex debugging, security-sensitive analysis, and long autonomous tasks where a failed run wastes far more developer time than the incremental inference cost. It is harder to justify for routine code completion, short documentation edits, simple test generation, or first-pass triage.
A sensible Copilot policy now has to distinguish between capability-critical jobs and high-volume convenience work. That means looking beyond the model’s benchmark score and measuring the unit economics that GitHub’s billing system now exposes:
But the reported CoreAI memo shows that model neutrality has limits inside Microsoft. The company is not merely brokering access to competing models; it has a long-term financial, licensing, infrastructure, and ownership relationship with one of them. OpenAI’s GPT-5.6 Sol is therefore both a model option and a strategic asset Microsoft has an incentive to utilize.
There is no evidence in CNBC’s reporting that Microsoft plans to impose the same default on external GitHub Copilot customers, remove rival models, or change enterprise administrators’ control over available models. GitHub’s public documentation still describes a model picker, plan-specific availability, organization-level enablement, and gradual rollout—not a mandatory Sol setting.
What Microsoft has not disclosed is more revealing than what it has. The company has not published the workload classes covered by Parikh’s instruction, the expected percentage of internal coding work moving to Sol, comparative quality results against Anthropic or Microsoft-developed coding models, or the actual effective cost difference under Microsoft’s OpenAI agreements. No other outlet has independently reported the memo’s timing or terms beyond CNBC’s account.
For Copilot administrators, the takeaway is narrower and more useful: Microsoft’s internal preference does not settle your model policy. It confirms that AI model selection has become a procurement and margin decision as much as a developer-experience decision. GitHub’s billing changes ensure that customers now have to make that same decision with their own budgets—not Microsoft’s OpenAI investment—on the line.
For developers outside Microsoft, the immediate practical change is none: GPT-5.6 Sol remains a selectable Copilot model where the relevant plan and organization policy permit it. GitHub’s own release documentation says Sol is intended for complex reasoning across large codebases and demanding, long-running agentic work; it calls GPT-5.6 Terra the balanced model for day-to-day interactive and agentic coding, and GPT-5.6 Luna the low-cost option for smaller, faster tasks.
That makes Parikh’s instruction a revealing internal economic decision, not a public declaration that every GitHub Copilot user should select Sol.
Microsoft Is Maximizing the Value of a Contract It Already Has
The memo arrives after Microsoft and OpenAI rewrote the terms of their partnership earlier in 2026. OpenAI said in April that Microsoft retains a license to OpenAI models and products through 2032, while Microsoft continues receiving revenue-share payments from OpenAI through 2030, subject to a cap. Microsoft also remains a major OpenAI shareholder.Those facts give Microsoft a reason to treat OpenAI usage differently from usage of Anthropic, Google, open-weight models, or even its own developing model families. A dollar spent on OpenAI inference is not economically identical to a dollar sent to an unrelated provider. Microsoft’s licensing rights let it build OpenAI technology into products such as Copilot, and its continuing financial arrangements mean growing OpenAI usage can have benefits beyond the immediate compute bill.
Microsoft’s 2025 annual filing stated that it holds rights to OpenAI intellectual property—including models and infrastructure—for integration into its products. The revised 2026 arrangement made the license non-exclusive, but did not remove it. Parikh’s reported direction reflects the operational implication: when a workload can be handled by an OpenAI model that Microsoft is already entitled to deploy and from which it can still receive economic value, sending that workload elsewhere may be harder to justify internally.
The company has simultaneously been building its own AI capabilities. Microsoft created the CoreAI – Platform and Tools organization under Parikh in January 2025 to unite developer-facing platforms including GitHub, Visual Studio, Visual Studio Code, Azure AI Foundry, and related engineering work. The existence of that division makes this memo more consequential than a routine employee recommendation. It concerns the groups that shape how many developers encounter AI tooling in practice.
Yet the direction should not be misread as Microsoft abandoning a multi-model approach. GitHub Copilot continues to expose choices from multiple model providers, and Microsoft’s public documentation for Copilot-based development workflows still recommends current Claude Sonnet, Claude Opus, OpenAI Codex, and GPT-5-class models depending on the task. Microsoft’s external posture remains model choice; the reported internal posture is to prefer the partner model when the economics support it.
GitHub’s Public Documentation Points to Terra, Not Sol, for Everyday Work
GitHub announced GPT-5.6 Sol, Terra, and Luna availability in Copilot on July 9, the same day OpenAI released the GPT-5.6 family. Its framing is unusually clear:- GPT-5.6 Sol is positioned for the highest reasoning ceiling, large-codebase analysis, and longer-running agent work.
- GPT-5.6 Terra is described by GitHub as the balanced default for everyday coding and agentic use.
- GPT-5.6 Luna is positioned as the faster, lower-cost member of the family.
The most plausible explanation is commercial, not technical. A company writing, reviewing, and maintaining the developer products themselves may value top-tier reasoning quality on internal projects more highly than it values per-task cost. But Parikh’s stated rationale, as reported by CNBC, was explicitly about extracting more value from Microsoft’s token investment in OpenAI. That is a business optimization layered over engineering preferences.
For an enterprise running GitHub Copilot, this is a useful warning against treating an employee memo as deployment guidance. A model that makes sense for Microsoft’s own internal economics can be a poor default for a customer whose Copilot bill is calculated from provider list pricing and whose simpler tasks dominate volume.
GitHub’s current model catalog does not make Sol universally available, either. Sol is offered on Copilot Pro+, Max, Business, and Enterprise plans, while Terra and Luna have broader availability that includes the standard Pro tier. For Business and Enterprise customers, GPT-5.6 access must be enabled by an administrator; GitHub says that policy is off by default. Availability is also rolling out gradually across Copilot surfaces.
So even if a CIO wanted to mirror Microsoft’s reported internal default, many organizations would first need to enable the model, confirm their plan supports it, and decide whether that is justified by their workload profile.
Usage-Based Billing Turns Model Choice Into a Budget Control
The timing also matters because GitHub Copilot moved to usage-based billing on June 1, 2026. GitHub AI Credits now reflect token consumption, including input, output, and cached tokens, at listed model API rates. Business and Enterprise organizations can pool included usage and set budgets at enterprise, cost-center, and individual-user levels.Under that structure, the model picker is no longer merely a quality setting. It is a spending control.
GitHub’s July release notice says GPT-5.6 models are billed at provider list pricing through Usage Based Billing. Microsoft has not publicly said whether its internal teams pay those same rates through internal chargeback systems, receive preferential accounting treatment, or have a separate arrangement for OpenAI capacity. That missing detail is central to the memo’s logic: the company’s claimed “greater value” may arise from partnership economics that a normal Copilot customer cannot reproduce.
Administrators should therefore resist a blunt policy of forcing every developer onto the biggest available model. Sol may be appropriate for repository-wide refactors, architecture reviews, complex debugging, security-sensitive analysis, and long autonomous tasks where a failed run wastes far more developer time than the incremental inference cost. It is harder to justify for routine code completion, short documentation edits, simple test generation, or first-pass triage.
A sensible Copilot policy now has to distinguish between capability-critical jobs and high-volume convenience work. That means looking beyond the model’s benchmark score and measuring the unit economics that GitHub’s billing system now exposes:
- Track which teams consume the largest share of input, output, and cached tokens before setting a model-wide default.
- Reserve the highest-reasoning models for workflows where reduced rework or fewer failed agent runs can be demonstrated.
- Use budget caps and cost-center controls before enabling expensive models across an enterprise organization.
- Test the same representative repositories and tasks across Sol, Terra, and lower-cost alternatives instead of assuming a frontier model wins on total cost per completed task.
The Memo Exposes the Limits of “Model Neutral” Copilot
Microsoft has spent considerable effort making GitHub Copilot a multi-model platform. Customers can select models from OpenAI, Anthropic, Google, and others, and the company’s product messaging increasingly describes model choice as task-dependent. That remains true at the product layer.But the reported CoreAI memo shows that model neutrality has limits inside Microsoft. The company is not merely brokering access to competing models; it has a long-term financial, licensing, infrastructure, and ownership relationship with one of them. OpenAI’s GPT-5.6 Sol is therefore both a model option and a strategic asset Microsoft has an incentive to utilize.
There is no evidence in CNBC’s reporting that Microsoft plans to impose the same default on external GitHub Copilot customers, remove rival models, or change enterprise administrators’ control over available models. GitHub’s public documentation still describes a model picker, plan-specific availability, organization-level enablement, and gradual rollout—not a mandatory Sol setting.
What Microsoft has not disclosed is more revealing than what it has. The company has not published the workload classes covered by Parikh’s instruction, the expected percentage of internal coding work moving to Sol, comparative quality results against Anthropic or Microsoft-developed coding models, or the actual effective cost difference under Microsoft’s OpenAI agreements. No other outlet has independently reported the memo’s timing or terms beyond CNBC’s account.
For Copilot administrators, the takeaway is narrower and more useful: Microsoft’s internal preference does not settle your model policy. It confirms that AI model selection has become a procurement and margin decision as much as a developer-experience decision. GitHub’s billing changes ensure that customers now have to make that same decision with their own budgets—not Microsoft’s OpenAI investment—on the line.
References
- Primary source: CNBC
Published: 2026-08-05T18:31:08+00:00
Microsoft AI exec tells developers to default to OpenAI's top model as part of efficiency push
Microsoft is taking advantage of its intellectual property rights with OpenAI, encouraging employees to use the company's top model.www.cnbc.com - Related coverage: openai.com
GPT-5.6: Frontier intelligence that scales with your ambition | OpenAI
More intelligence from every token, stronger performance per dollar, and more capability on demand for your hardest work.
openai.com
- Related coverage: help.openai.com
Model Release Notes | OpenAI Help Center
help.openai.com
- Related coverage: openai.com
Previewing GPT-5.6 Sol: a next-generation model | OpenAI
OpenAI previews GPT-5.6 Sol, a next-generation model with stronger capabilities in coding, science, and cybersecurity, paired with its most advanced safety stack.openai.com - Related coverage: community.openai.com
Introducing GPT-5.6 series: Sol, Terra and Luna. Coming July 9 10am PT - Announcements - OpenAI Developer Community
Today OpenAI is previewing GPT-5.6 Sol, its newest flagship model for developers and enterprises, alongside Terra and Luna. Sol is built for frontier reasoning and long-horizon agentic work; Terra is a balanced every…
community.openai.com
- Related coverage: help.openai.com
GPT-5.6 in ChatGPT | OpenAI Help Center
Learn how GPT-5.6 Sol works in ChatGPT, which options are available by plan, and how usage limits and availability work.
help.openai.com
- Related coverage: developers.openai.com
GPT-5.6 Sol Model | OpenAI API
developers.openai.com
- Related coverage: deploymentsafety.openai.com
GPT-5.6 System Card - OpenAI Deployment Safety Hub
GPT-5.6 is a new family of three models: Sol, our new flagship model; Terra, a capable lower-cost option; and Luna, our fastest and most cost-efficient model. The safeguards we have built for this launch—our most robust yet—are built to deliver these models safely and at scale, around the world.deploymentsafety.openai.com - Related coverage: info.microsoft.com
- Related coverage: wsgr.com
- Related coverage: techxplore.com
- Related coverage: learn.microsoft.com
Step 2: Install GitHub Copilot Chat | Microsoft Learn
Install the GitHub Copilot Chat extension in Visual Studio Code, sign in, and pick a model that works reliably with the Employee Self-Service Agent Developer Kit.learn.microsoft.com - Related coverage: community.openai.com
Codex Throughput Scaling for Heavy AI Builder Workloads - Codex - OpenAI Developer Community
I’m building multi-repo AI systems using Codex in VSCode with frequent high-tier reasoning calls. My workload includes agent orchestration, backend services, data pipelines, and iterative architectural refactoring across…
community.openai.com
- Related coverage: kucoin.com
Microsoft Shifts AI Workloads in Excel and Outlook to In-House MAI Models | KuCoin
Microsoft has quietly started swapping out the AI brains behind Excel and Outlook. As of July 7, 2026, the company began routing a meaningful share of Copilot pwww.kucoin.com - Related coverage: learn.microsoft.com
- Related coverage: blogs.microsoft.com
Microsoft and OpenAI evolve partnership to drive the next phase of AI - The Official Microsoft Blog
We are thrilled to continue our strategic partnership with OpenAI and to partner on Stargate. Today’s announcement is complementary to what our two companies have been working on together since 2019. The key elements of our partnership remain in place for the duration of our contract through...blogs.microsoft.com - Related coverage: blogs.microsoft.com
Introducing CoreAI – Platform and Tools - The Official Microsoft Blog
Satya Nadella, Chairman and CEO, shared the below communication with Microsoft employees this morning. As we begin the new year, it’s clear that we’re entering the next innings of this AI platform shift. 2025 will be about model-forward applications that reshape all application categories. More...blogs.microsoft.com - Related coverage: news.microsoft.com
- Related coverage: microsoft.com
Code|AI - Microsoft Research
The Code!AI team in Microsoft’s CoreAI Division is shaping the future of AI for developers. We are a diverse, entrepreneurial, and multidisciplinary team of scientists, researchers, and engineers driven by the shared mission of empowering every developer and their team to achieve more using...www.microsoft.com - Related coverage: microsoft.gcs-web.com
- Related coverage: cdn.openai.com
- Related coverage: windowscentral.com
"The work we’re doing together remains ambitious": OpenAI ends Microsoft exclusivity in a tone that doesn’t match last week’s leaked memo | Windows Central
OpenAI’s official announcement sounds optimistic, but last week’s leaked memo told a different story.www.windowscentral.com - Related coverage: techradar.com
OpenAI and Microsoft make a huge change to their exclusive partnership - does this open the door for AWS to swoop in? | TechRadar
Microsoft is no longer OpenAI's exclusive cloud partnerwww.techradar.com - Related coverage: tomshardware.com
Microsoft and OpenAI end exclusivity agreement, opening up potential partnerships with Amazon and Google — Microsoft will continue to receive revenue share through 2030 | Tom's Hardware
OpenAI has broken Microsoft's containment.www.tomshardware.com - Related coverage: time.com
Why Elon Musk Is Suing OpenAI and Sam Altman
The lawsuit makes a substantial claim that could fundamentally reshape how OpenAI and its partner Microsoft do business.time.com
