Frontier Intelligence Directory · Updated May 20, 2026
The decision layer on top of the raw data. Every frontier provider, model, and agentic platform — categorized by capability, priced live, and paired with a verdict. Built for humans and agents.
We cite OpenRouter, Artificial Analysis, and LMArena as sources, and add what they don’t: task-first navigation, the agentic-platform comparison, curated verdicts, and a creator-stack lens.
5
Providers tracked
8
Frontier models
0
Agentic platforms
6
Live-priced
The fastest path from “which model?” to an answer. One dominant constraint → a recommendation.
| If you need… | Pick | Runner-up | Why |
|---|---|---|---|
| Lowest cost (closed) | gemini-3-5-flash | Claude Haiku 4.5 | Frontier agentic performance at less than half the cost of comparable flagships. |
| Lowest cost (open weights) | deepseek-v3-2 | qwen3-coder-next | Frontier-class reasoning at fraction-of-a-cent economics under MIT license. |
| Hardest reasoning | Claude Opus 4.6 | GPT-5.2 Pro | #1 ARC-AGI-2 (68.8%) and OSWorld (72.7%). |
| Agentic coding | gemini-3-5-flash | Claude Opus 4.6 | 76.2% Terminal-Bench 2.1 and 83.6% MCP Atlas at low cost. |
| Longest context | Grok 4.1 | Gemini 3 Pro | 2M-token native context with aggressive pricing. |
| Native voice | GPT-5.2 Pro | — | Native audio modality with no close runner-up. |
| Multimodal understanding | Gemini 3 Pro | GPT-5.2 Pro | Widest modality support; 81% MMMU-Pro. |
| Video generation | gemini-omni | — | Native frontier video gen with natural-language editing. |
| EU data sovereignty | mistral-large-3 | command-a-reasoning | Paris-based, full EU residency; Apache-licensed small-model pairing. |
| Self-host / own the weights | Llama 4 Maverick | deepseek-v3-2 | Open-weight MoE (400B/17B) that runs on a single H100. |
Pick the job, jump to the providers that lead.
Complex problem-solving, math, abstract reasoning, long-horizon planning
Vision, document, chart, and cross-modal reasoning across text/image/audio
Generative video models, text-to-video, image-to-video, editing
Agentic coding, terminal use, debugging, multi-file refactors
Tool use, function calling, agent SDKs, computer use, long-horizon execution
Native speech in/out, real-time conversation, audio understanding
Text-to-image, editing, in-painting, brand-consistent generation
Sort and filter every tracked model. Live pricing via OpenRouter where available. Click a model for the full breakdown.
8 models live pricing via OpenRouter
| Claude Opus 4.6Anthropic | 2026-02-05 | 1M | $5.00 | $25.00 |
| GPT-5.2 ProOpenAI | 2026-01-01 | 400K | $21.00 | $168.00 |
| Gemini 3 ProGoogle DeepMind | 2025-12-01 | 2M | — | — |
| Llama 4 MaverickMeta AI | 2025-12-01 | 1.0M | $0.15 | $0.60 |
| Claude Opus 4.5Anthropic | 2025-11-01 | 200K | $5.00 | $25.00 |
| Grok 4.1xAI | 2025-11-01 | 2M | — | — |
| Claude Haiku 4.5Anthropic | 2025-10-01 | 200K | $1.00 | $5.00 |
| Claude Sonnet 4.5Anthropic | 2025-09-29 | 1M | $3.00 | $15.00 |
Creators don’t pick a model — they assemble a stack. Here’s what to use across each modality, and the workflow.
Generate, arrange, and master AI music
Workflow. Draft lyrics + style prompt with Claude → generate in Suno → iterate sections → export and master. The LLM is the creative director; Suno is the band.
AI Music MasterclassGenerate and edit on-brand visuals
Workflow. Concept + prompt with an LLM → generate hero with Imagen/Nano Banana → object-level edits (swap, resize, recolor) → export to the design system.
Visual creation systemGenerate and edit video with natural language
Workflow. Script with an LLM → generate with Omni → edit by instruction (background swap, camera angle) → produce. Increasingly a single agentic sequence.
Long-form, on-voice writing at depth
Workflow. Research + outline with Opus (1M context holds your whole corpus) → draft → tighten with Sonnet → publish. The brand voice gate stays human.
Content StudioShip code with agentic assistants
Workflow. Flash for the high-volume agent loop, Opus 4.6 for the critical reasoning path. Run inside Claude Code, Cursor, or Antigravity 2.0.
Frontier Models ArenaWhere the models actually do work — IDEs, CLIs, desktops, agent platforms, managed runtimes. The layer the data sites skip.
We synthesize; we don’t fabricate. Live pricing is attributed; benchmarks are sourced; vendor-reported figures are labelled as such and flagged pending independent reproduction.
OpenRouter
Live per-token pricing & availability (300+ models)
Artificial Analysis
Independent Intelligence Index, speed, latency
LMArena
Crowdsourced human-preference Elo
ARC Prize Foundation
ARC-AGI abstract reasoning benchmark
SWE-bench
Real-world software engineering tasks
Vendor model cards
Self-reported benchmarks (labelled as such)
There is no single winner. Claude Opus 4.6 leads reasoning (68.8% ARC-AGI-2) and agentic coding. GPT-5.2 Pro dominates broad multimodal + voice. Gemini 3.5 Flash (Google I/O ’26) sets a new cost/intelligence frontier at less than half the cost of comparable flagships. Gemini 3.5 Pro ships next month for the highest-tier reasoning. Pick by task — use the decision matrix above.
Those are the raw-data sources — OpenRouter for live pricing and routing, Artificial Analysis for independent benchmarks, LMArena for human preference. We cite all three. The FrankX LLM Hub adds the decision layer they don’t: task-first navigation, the agentic-platform comparison (Claude Code vs Antigravity vs Cursor vs Codex), curated verdicts, and a creator-stack lens — for humans and agents.
DeepSeek V3.2 leads on pure cost ($0.27 / $1.10 per 1M tokens, MIT license). Gemini 3.5 Flash is the cheapest closed-frontier option at $0.30 / $2.50. Both deliver frontier-class reasoning for production agentic workloads.
By category: coding agents — Gemini 3.5 Flash (76.2% Terminal-Bench 2.1) and Claude Opus 4.6; long-horizon enterprise — Gemini Spark and Claude Agent Teams; computer-use — GPT-5.2 Operator and Claude Opus 4.6 (72.7% OSWorld).
Where a model maps to OpenRouter, pricing is fetched live (hourly) and marked with a ⚡ icon and "via OpenRouter." Otherwise it comes from our curated registry. Always verify against the provider before relying on it for billing.
Yes. The full curated dataset — models, pricing, verdicts, decision matrix, comparisons — is available as clean JSON at /llm-hub.json, plus JSON-LD structured data on every page and deep links in /llms.txt.