Refreshed August 2, 2026 · 18 models

Every frontier model. One place.

Compare 18 production models across OpenAI, Anthropic, Google, xAI, Moonshot, Alibaba, Z.ai, Xiaomi, MiniMax, and DeepSeek — on benchmarks, pricing, context, and practical task fit. Pick a model in 30 seconds, or evaluate a real task in two minutes.

Pick a model in 30 seconds Browse all models

Jul · New line GPT-5.6 Sol / Terra / Luna OpenAI's new three-tier line. Sol is the 1M-context flagship at $5/$30; Terra ($2/$12) and Luna ($0.20/$1.20) reset the value tiers.

Jul · New GA Claude Opus 5 + Sonnet 5 Opus 5 is a drop-in upgrade over Opus 4.8 at the same $5/$25. Sonnet 5 brings near-Opus coding at $3/$15 (intro $2/$10 through Aug 31).

Jul 27 · OSS Kimi K3 (Moonshot) 2.8T-param MoE — the largest open model ever. New open-weight intelligence leader (AA Index 57 vs GLM 5.2's 51); #1 on Arena Frontend Code.

Quick pick

Three questions — one recommended model. For nuanced tradeoffs, use the full evaluator below.

1. What's the work?

2. Budget?

3. Speed need?

The full lineup

Click any card to see strengths, watch-outs, and tooling. Pin up to 3 to compare side-by-side.

Task evaluator

Paste a real task, choose the dominant work type, and tune cost, speed, and context sensitivity. Scored against all 18 production models.

Your task

Primary task type

Approximate input size

Cost sensitivity

Latency sensitivity

Context importance

Tool-use importance

Recommendation

Scores fit on a 100-point scale across all production models. Top 5 shown with tradeoff rationale.

Side-by-side prompt runners

Top 3 recommended models get starter prompts. Paste into vendor consoles or OpenRouter and score outputs consistently.

Scorecards

Repeatable rubric for head-to-head runs.

Best team of models

Splits a workflow into subtasks and assigns the best model for each slice — routing across the full 18-model set.

Multi-step workflow

Routed plan

Opinionated defaults — override freely.

Ones to watch

Models not yet production-stable for international use, or restricted access only — but worth tracking closely.

How to use this in practice

Start with the evaluator for a quick pick. For zero-cost experimentation, try Qwen3.6-Plus, MiniMax M3, or GLM 5.2 on OpenRouter before paying for frontier models. Use the router for compound workflows. Check the "Ones to watch" section for models that may leapfrog the current leaders — Kimi K3 just took the open-weight crown from GLM 5.2, Claude Sonnet 5 undercut its own Opus tier, and GPT-5.6 Luna reset the budget lane. The gap between open and closed is the narrowest it has ever been.