The gap between closed and open AI models just became a chasm, and the companies with the compute budgets are sprinting ahead.

The Summary

The Signal

Artificial Analysis updated its Intelligence Index with a new methodology that crowns the companies everyone expected: Google, Anthropic, and OpenAI. Gemini 4 Carbon sits at the top with a score of 59, just ahead of Claude Opus 5.5 and GPT-6 Astra. The margins are tight, but the message is clear. If you want frontier intelligence, you're buying it from one of three vendors.

The index isn't just a vanity metric. It aggregates performance across reasoning, coding, multimodal tasks, and mathematical problem-solving. These are the capabilities that matter for autonomous agents that need to navigate complex workflows without human handholding. A model that scores 59 versus 54 might be the difference between an agent that completes a task and one that gets stuck asking for clarification.

"The updated Intelligence Index highlights the growing divide between proprietary and open-weight AI models, impacting innovation dynamics."

The launch of the HLE-Diamond leaderboard alongside the index reveals what's really happening. Proprietary models are getting smarter faster than open alternatives can keep pace. Meta's Llama derivatives and Mistral's open releases serve specific use cases well, but they're not closing the intelligence gap. They're optimizing for different tradeoffs: cost, control, customization. Meanwhile, Google, Anthropic, and OpenAI are optimizing for raw capability, consequences be damned.

This matters for anyone building in the agent economy. If you're architecting a system that needs frontier reasoning, you're renting compute from one of the big three. If you're building something that needs to run locally, operate at lower cost, or avoid vendor lock-in, you're accepting an intelligence ceiling. The index makes these tradeoffs quantifiable, which means strategic decisions just got clearer and harder at the same time.

Key competitive dynamics:

  • Google, Anthropic, and OpenAI control the frontier, with scores clustered between 54-59
  • Open-weight models lag by 10-15 points on intelligence metrics, a gap that's widening
  • ARK Invest is tracking these rankings as a signal for where capital flows next in AI infrastructure

The Implication

If you're deploying agents in 2027, you need to decide whether you're optimizing for intelligence or independence. Frontier models give you the smartest workers. Open models give you workers you own. The index just made that choice explicit and measurable.

Watch what happens to developer tooling next. The companies that figure out how to make open models punch above their weight on specific domains will carve out valuable niches. But the general-purpose intelligence crown stays with the big three until compute costs collapse or someone finds a fundamentally different architecture. Neither is happening tomorrow.

Sources

Crypto Briefing