GitHub to charge Copilot by usage as inference costs rise

GitHub announced that Copilot will soon be billed according to actual AI usage. The change reflects a refusal to shoulder ever‑growing inference costs for its most demanding users.

The company said it can no longer absorb the “escalating inference cost” from its heaviest AI users. Copilot, the code‑completion tool that powers millions of daily suggestions, will switch from a flat‑rate subscription to a metered model. Existing plans remain for light users, but heavy teams will see usage‑based charges on their next invoice.

GitHub shifts to usage‑based billing

GitHub’s statement linked the pricing shift to raw compute expenses. The service runs large language models in the cloud, and each token generated consumes GPU time. As model size and request volume increase, the underlying hardware bill climbs faster than subscription fees.

The move puts GitHub in line with other AI platform providers that already charge per request. Microsoft Azure, Google Cloud, and AWS all expose inference costs to customers. GitHub’s decision signals that the free‑tier subsidy model is unsustainable for a product that now processes billions of tokens per month.

Developers who rely on Copilot for daily coding will need to audit their usage. Teams can monitor token consumption in the GitHub dashboard and set alerts to avoid surprise charges. The change also forces enterprises to evaluate the ROI of AI‑assisted development versus traditional tooling.

Anthropic backs Blender’s open‑source roadmap

Anthropic joined the Blender Development Fund as a corporate patron. The fund, administered by the Blender Foundation, supports core development of the open‑source 3D creation suite. Anthropic’s contribution marks its first formal sponsorship of a non‑AI open‑source project.

The partnership appears motivated by shared interests in community‑driven software. Anthropic, known for its large‑scale language models, gains visibility among creators who increasingly embed AI tools into visual pipelines. Blender, in turn, receives a steady revenue stream that can accelerate feature work and stability fixes.

Industry observers note that AI companies are looking beyond pure model research. By funding creative tools, they tap into a user base that may later adopt AI plugins, extensions, or cloud services. The patronage also signals a willingness to support open ecosystems rather than rely solely on proprietary platforms.

Open Benchmarks Grants aim to close evaluation gap

Snorkel announced a $3 million commitment to launch Open Benchmarks Grants. The program will fund research and development of open benchmarks that measure AI performance in realistic settings. Partners include Hugging Face, Prime Intellect, Together AI, Factory HQ, Harbor, and PyTorch.

The grant call targets the growing asymmetry between model capability and evaluation rigor. Benchmarks such as Terminal‑Bench, METR, and ARC‑AGI have highlighted gaps in current testing regimes, especially for agentic systems. Open Benchmarks Grants will prioritize environments that capture real‑world complexity, longer autonomy horizons, and richer output criteria.

Selected teams will receive cash awards, expert data‑development support, and access to shared infrastructure. The initiative acknowledges that academic and open‑source groups often lack the resources to build large‑scale evaluation suites. By plugging that funding hole, Snorkel hopes to reduce the risk of “benchmaxxing,” where models overfit narrow test sets without real‑world robustness.

Industry pressure and the economics of AI inference

GitHub’s pricing shift, Anthropic’s patronage, and Snorkel’s grant program all reflect a broader economic pressure on AI services. Inference—running a model to generate output—remains the dominant cost driver for cloud AI providers. As models grow from hundreds of millions to tens of billions of parameters, each query consumes more GPU cycles.

Companies that once offered free or flat‑rate access now face a sustainability dilemma. The Copilot announcement shows that even well‑capitalized subsidiaries of tech giants must pass costs downstream. For developers, the implication is clear: AI assistance will increasingly be treated as a utility, measured in compute units rather than a feature.

At the same time, the AI community is investing in open‑source infrastructure to mitigate lock‑in risk. Anthropic’s sponsorship of Blender signals that AI firms see value in nurturing ecosystems where third‑party tools can flourish. Open benchmarks, funded by Snorkel, aim to provide neutral metrics that prevent proprietary claims of superiority.

These parallel tracks suggest a market where AI providers monetize raw compute while supporting open standards that keep the ecosystem competitive. The tension between revenue generation and community health will shape product roadmaps for the next few years.

What to watch

Watch GitHub’s billing dashboard for the first usage‑based invoices, which will reveal average token consumption per developer. Track Anthropic’s announcements for any AI‑centric plugins or services built on top of Blender. Monitor the Open Benchmarks Grants awardees and the first public benchmark releases they produce. Together, these signals will indicate how quickly the AI stack is moving from experimental freebies to a commoditized, usage‑priced infrastructure.