
§ 00 MODEL CATALOG · ARENA REFERENCE
Rankings across text, image, vision, video, and agents. Route frontier models through one EU OpenAI-compatible API. Free playground with a PII vault — not a vendor :free SKU that may train on prompts.
On this board
§ 01 Arena Leaderboard
Switch this table from the workspace sidebar — Text, Image, Video, Vision, Agents, or €/M. Open full window for the same board without page chrome. Compare gateway-routable frontier models or projected sovereign throughput.
| # | Model | Arena Win % | Provider | List price (in/out) | Gateway (SRA intro) | Actions |
|---|---|---|---|---|---|---|
| 1 | Claude Fable 5 claude-fable-5· gateway | 58.2% Arena win rate (reference) | 🇺🇸 Anthropic | $5.00 / $25.00 | €0.10/M / €0.10/M | |
| 2 | Claude Opus 4.8 Thinking claude-opus-4-8-thinking· gateway | 57.1% Arena win rate (reference) | 🇺🇸 Anthropic | $5.00 / $25.00 | €0.10/M / €0.10/M | |
| 3 | GPT-5.5 xHigh gpt-5.5-xhigh· gateway | 56.4% Arena win rate (reference) | 🇺🇸 OpenAI | $1.75 / $14.00 | €0.10/M / €0.10/M | |
| 4 | Claude Opus 4.8 claude-opus-4-8· gateway | 55.8% Arena win rate (reference) | 🇺🇸 Anthropic | $5.00 / $25.00 | €0.10/M / €0.10/M | |
| 5 | Claude Opus 4.6 Thinking claude-opus-4-6-thinking· gateway | 55.2% Arena win rate (reference) | 🇺🇸 Anthropic | $5.00 / $25.00 | €0.10/M / €0.10/M | |
| 6 | Gemini 3.1 Pro Preview gemini-3.1-pro-preview· gateway | 54.6% Arena win rate (reference) | $1.50 / $12.00 | €0.10/M / €0.10/M | ||
| 7 | GPT-5.5 High gpt-5.5-high· gateway | 54.0% Arena win rate (reference) | 🇺🇸 OpenAI | $1.75 / $14.00 | €0.10/M / €0.10/M | |
| 8 | Claude Sonnet 5 High claude-sonnet-5-high· gateway | 53.4% Arena win rate (reference) | 🇺🇸 Anthropic | $2.00 / $10.00 | €0.10/M / €0.10/M | |
| 9 | Grok 4.5 grok-4.5· gateway | 52.8% Arena win rate (reference) | 🇺🇸 xAI | $1.25 / $2.50 | €0.10/M / €0.10/M | |
| 10 | DeepSeek V4 Pro deepseek-v4-pro· gateway | 52.1% Arena win rate (reference) | 🇨🇳 DeepSeek | $0.60 / $1.70 | €0.10/M / €0.10/M |
Rankings on this page are reference data from LMSYS Chatbot Arena (arena.ai) as of the date shown. AGICY does not run arena evaluations and does not claim independent benchmark results. Sovereign fleet throughput is modeled; gateway prices are AGICY list rates (€0.10/M in + €0.10/M out), not official provider pricing.
Arena reference: scrape-first ingest (12h CI) writes leaderboard-live.json; this catalog is a fallback / SEO index. AGICY does not run LMSYS Arena evals. Text top mirror as-of 2026-07-28. Grok 4.6 / GLM-5.2 / Kimi K3 / Qwen3.8 / MiniMax-M3 rows from live snapshot asOf 2026-08-12. Ingest: python .archon/scripts/calc_leaderboard_ingest.py — scrape-first, no paid AA API key. 14h wake-up via GitHub Action + /api/leaderboard/refresh. Snapshot hash 7681e7763a. arena.ai/leaderboard →
Consensus is a normalized, freshness-weighted blend of scraped public boards (Arena, AA, HF Open LLM, OpenRouter usage) for gateway routing guidance. It is NOT LMSYS Arena ELO. OpenRouter contributes popularity weight only.
§ 02 Tenstorrent Performance
1080p, 5-second, 24fps video generation time. Lower is better. Tenstorrent hardware delivers 3.5 seconds — verified by Artificial Analysis.
§ 04 Sovereignty Shield
PII can be vaulted before US upstream hops. Residency is not sovereignty. The Vasilikos campus is pre-construction.
Physical network isolation. Data Embassy configuration. Hardware you can touch.
Tenstorrent Blackhole RISC-V architecture eliminates CUDA lock-in. 5× less per server vs $3.7M for DGX B200.
GDPR, NIS2, EU AI Act design alignment; SOC 2 and ISO 27001 are roadmap targets — designed in, not claimed as held.
§ 05 Cost Efficiency
| Tier | Retail Benchmark | ||
|---|---|---|---|
e.g. Mistral Small 4 | €0.20–2.00/M | ~€0.14/M | €0.10/M (SRA intro) |
e.g. GLM-5.2 | €1.50–5.00/M | ~€1.33/M | €0.10/M (SRA intro) |
e.g. Kimi K2.6 | €2.00–15.00/M | ~€1.57/M | €0.10/M (SRA intro) |
| Data Sovereignty | ❌ US | ❌ Public | ✅ EU |
Copperway MultiPass routes many families — frontier and open-weight — through one EU API. Openness is the catalog, not a single SKU.