The Best AI Models Right Now (July 2026)

A dated snapshot of every leaderboard: the smartest, fastest, cheapest and best-for-X models as of July 2026.

This is a dated snapshot of modelgrep's live rankings as of July 24, 2026. The linked pages update hourly; this post records where the field stood mid-2026 — useful for tracking how fast the leaderboard turns over.

Smartest overall

Claude Fable 5 leads the Intelligence Index ranking at 59.9, ahead of GPT-5.6 Sol (58.9) and Kimi K3 (57.1). The top three span three different labs — a notably tighter race than the start of the year.

Best for coding

GPT-5.6 Sol tops the Coding Index at 77.4, with GPT-5.6 Terra (76.7) and Claude Fable 5 (76.5) within a point. At this spread, price and latency should decide your pick — see best models for Cursor and for Claude Code.

Best for design & frontend

Kimi K3 holds a commanding Design Arena Elo lead at 1464 — nearly 100 Elo clear of Claude Fable 5 (1370) and GLM 5.2 (1358). Human preference on rendered UI remains the benchmark general intelligence scores predict worst.

Fastest and cheapest

Gemini 3.5 Flash is the fastest model at 159 tokens/sec. On price, Ling-2.6-flash serves at $0.010 per million input tokens, with Granite 4.0 Micro ($0.017) and Mistral Nemo ($0.019) close behind — the budget tier is now effectively free at prototyping volumes (run your own numbers).

Open source & local

GLM 5.2 is the strongest open-weight model at 51.1 intelligence — frontier-class and self-hostable, though far too large for consumer hardware. Among models you can actually run locally, MiniMax M3 (44.4) leads; see the 16GB RAM and 32GB RAM tiers for what fits your machine.

These rankings have probably changed

Every page linked above refreshes hourly with live benchmarks, speed and pricing.

See the current leaders

What changed since January

The intelligence race tightened. Under 3 points now separate the top three labs.
Coding scores converged at the top — the leaders sit within a single point, so operational metrics (speed, price, caching) matter more than the benchmark delta.
Open weights crossed 50 intelligence. GLM 5.2 put frontier-class capability on Hugging Face.
Design stayed winner-take-most. Kimi K3's ~100-Elo lead on UI generation is the largest gap on any leaderboard.