Eighteen months ago, GPT-4 was clearly the best large language model. Nothing else came close on most benchmarks.

Today, the gap between OpenAI’s latest and alternatives—Claude, Gemini, Llama, Mistral, and others—has narrowed dramatically. For many business applications, the choice between models is increasingly about price, features, and vendor relationship rather than fundamental capability differences.

We’re watching AI models commoditize faster than almost anyone predicted.

The Evidence

Consider benchmark trajectories:

Coding tasks: Top open-source models now match or exceed proprietary models from 12 months earlier on standard coding benchmarks.

General reasoning: The gap between tier-one closed models (GPT-4, Claude Opus, Gemini Ultra) and tier-two options has compressed from substantial to marginal.

Specialised domains: Fine-tuned smaller models often outperform general frontier models on specific tasks, at fraction of the cost.

Price movements tell the same story. API costs for comparable capability have dropped roughly 90% since early 2023. Competition is driving prices toward marginal cost of compute.

Why This Matters

Commoditization has strategic implications across the AI value chain:

For AI Companies

The moat-building playbook is under pressure:

Data advantages erode quickly. Training data approaches that produce capability jumps get replicated within quarters. Synthetic data and better training techniques spread rapidly through the research community.

Scale advantages hit limits. Throwing more compute at larger models produces diminishing returns. The paradigm of “make it bigger” appears to be plateauing.

Differentiation shifts downstream. If base model capability is comparable, competition moves to fine-tuning, tooling, enterprise features, and customer success.

OpenAI’s shift toward platform services (GPTs, assistants, enterprise features) reflects this reality. Raw model capability alone doesn’t sustain premium pricing indefinitely.

For Businesses Using AI

Commoditization creates both opportunities and risks:

Opportunities:

• Falling costs make AI economical for more use cases
• Multi-vendor strategies become practical
• Reduced dependence on any single provider

Risks:

• Difficulty building sustainable differentiation on AI capabilities competitors can also access
• Investment in specific model integrations may become stranded
• “AI-powered” as a competitive advantage diminishes

The strategic question shifts from “should we use AI?” to “how do we create value that persists as AI becomes commoditized infrastructure?”

For AI Strategy

Building on commoditizing infrastructure requires different approaches:

Avoid tight coupling. Abstraction layers that allow model switching protect against vendor lock-in and enable cost optimisation.

Focus on data and workflow. Proprietary data, domain-specific fine-tuning, and integration with business processes create more durable advantages than base model selection.

Plan for falling prices. Today’s expensive AI applications become economically viable as costs drop. Pipeline projects accordingly.

Monitor capability convergence. Evaluate whether premium models justify premium prices for your specific use cases. The answer changes over time.

What’s Not Commoditizing

Some dimensions of AI remain differentiated:

Speed and reliability. Production systems need consistent latency and uptime. Provider operations and infrastructure vary significantly.

Enterprise features. Security certifications, compliance, audit logs, fine-grained access control—these matter to enterprise buyers regardless of model parity.

Ecosystem and tooling. Developer experience, documentation, integration options, and support quality create meaningful differences.

Specific capabilities. Multimodal processing, long context handling, agentic features—niche capabilities remain unevenly distributed.

Trust and relationship. For major deployments, vendor stability, responsiveness, and alignment with business needs matter beyond pure technical capability.

The Open Source Wild Card

Open-source models complicate the commoditization picture:

Meta’s Llama series, Mistral’s releases, and community-developed models provide alternatives to commercial APIs. For organisations with technical capacity, running open models offers:

• Cost control (pay for compute, not API margins)
• Data privacy (processing stays internal)
• Customisation (unrestricted fine-tuning)

The capability gap between open and closed models continues narrowing. Some analysts expect convergence within 2-3 years for most commercial applications.

This doesn’t mean everyone should run their own models. Operational complexity is real. But it adds competitive pressure to commercial pricing.

Strategic Implications

For businesses developing AI strategy:

Don’t over-invest in model-dependent architectures. Build flexibility to switch between providers and approaches.

Invest in data and workflows. Your proprietary data, domain expertise, and integrated workflows are more defensible than your model choice.

Think about AI as infrastructure. Like cloud computing before it, AI is transitioning from differentiator to baseline capability.

Watch the cost curve. Applications uneconomical today may be viable within 12-24 months as prices continue falling.

Consider timing. If you’re building AI-centric products, understand that capability advantages based on model access are temporary.

The Broader Picture

AI model commoditization reflects a familiar technology pattern:

1. Innovation creates new capability
2. Early movers capture premium value
3. Competition and imitation compress advantages
4. Value migrates to applications and ecosystems
5. Capability becomes infrastructure

We’re somewhere between stages 3 and 4. The AI companies that thrive long-term will likely be those building ecosystems and applications rather than solely racing on model capability.

For businesses using AI, the implication is clear: the strategic question isn’t which model to pick. It’s how to build lasting value on top of infrastructure that’s becoming a commodity.