SFX-10 INDEX
858.4
▲ 1.90%
+16.0 pts
Composite index of the top 10 frontier AI models. Score = tape (55%) + fundamentals (30%) + cost efficiency (15%), with trust and freshness penalties applied before each source enters the blend. Refreshed every 6 hours.
Updated 7 Oct 2026, 5:03 pm AEST
CL-SN55GPT-61SGPT-56TGPT-6SGPT-56SCCL-OP55XGROK-46HGLM-52DSK-V4PHGLM-5
All Models
30 models tracked| # | Symbol | Model | Vendor | Score | Text ELO | Code ELO | Tape | Fund | Cost Idx | $/M in | $/M out | Ctx | Conf |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | CL-SN55 SFX-10 | Claude Sonnet 5.5 | Anthropic | 941 | 1715 | 1715 | 1.000 | 0.900 | 0.803 | $2.00 | $10.00 | 1M | ★★☆ |
| 2 | GPT-61S SFX-10 | GPT-6.1 Sol | OpenAI | 890 | 1758 | 1758 | 0.959 | 0.808 | 0.803 | $2.00 | $10.00 | 1M | ★★☆ |
| 3 | GPT-56T SFX-10 | GPT-5.6 Terra | OpenAI | 861 | — | — | 0.962 | 0.798 | 0.620 | — | — | — | ★★☆ |
| 4 | GPT-6S SFX-10 | GPT-6 Sol | OpenAI | 847 | 1689 | 1689 | 1.000 | 0.589 | 0.803 | $2.00 | $10.00 | 1M | ★★☆ |
| 5 | GPT-56SC SFX-10 | GPT-5.6 Sol Codex Harness | OpenAI | 847 | 1620 | 1620 | 1.000 | 0.589 | 0.803 | $2.00 | $10.00 | 1M | ★★☆ |
| 6 | CL-OP55X SFX-10 | claude-opus-5.5-max | Anthropic | 843 | 1815 | 1815 | 1.000 | 0.589 | 0.778 | — | — | — | ★☆☆ |
| 7 | GROK-46H SFX-10 | grok-4.6-high | SpaceXAI | 843 | 1620 | 1620 | 1.000 | 0.589 | 0.778 | — | — | — | ★☆☆ |
| 8 | GLM-52 SFX-10 | GLM-5.2 | Z.ai | 843 | 1605 | 1605 | 1.000 | 0.589 | 0.778 | $0.07 | $12.00 | 1M | ★★☆ |
| 9 | DSK-V4PH SFX-10 | deepseek-v4-pro-high-20260813 | DeepSeek | 843 | 1583 | 1583 | 1.000 | 0.589 | 0.778 | — | — | — | ★☆☆ |
| 10 | GLM-5 SFX-10 | GLM-5 | Z.ai | 826 | 1616 | 1616 | 0.887 | 0.646 | 0.965 | $0.60 | $1.92 | 205K | ★★☆ |
| 11 | MUSE-SPK | Muse Spark | Meta | 804 | 1489 | 1624 | 0.799 | 0.715 | 1.000 | $0.10 | $0.20 | 1M | ★★☆ |
| 12 | QWEN-38X | Qwen3.8-Max | Alibaba | 795 | 1670 | 1670 | 0.887 | 0.646 | 0.753 | $4.00 | $12.00 | 1M | ★★☆ |
| 13 | GPT-6A | GPT-6 Astra | OpenAI | 782 | 1788 | 1788 | 0.969 | 0.831 | 0.000 | $10.00 | $50.00 | 1M | ★★☆ |
| 14 | CL-FAB5H | claude-fable-5-high | Anthropic | 761 | 1504 | 1625 | 0.851 | 0.589 | 0.778 | — | — | — | ★☆☆ |
| 15 | CL-OP5H | Claude Opus 5 (high) | Anthropic | 736 | 1490 | 1660 | 0.805 | 0.589 | 0.778 | — | — | — | ★☆☆ |
| 16 | CL-OP5X | Claude Opus 5 (max) | Anthropic | 734 | 1489 | 1695 | 0.802 | 0.589 | 0.778 | — | — | — | ★☆☆ |
| 17 | GEM-37FH | Gemini 3.7 Flash High | 732 | 1488 | 1592 | 0.798 | 0.589 | 0.778 | — | — | — | ★☆☆ | |
| 18 | KIMI-K3X | Kimi K3 Max | Moonshot | 719 | 1488 | 1658 | 0.756 | 0.623 | 0.778 | — | — | — | ★☆☆ |
| 19 | CL-OP46H | Claude Opus 4.6 High | Anthropic | 719 | 1505 | — | 0.775 | 0.589 | 0.778 | — | — | — | ★☆☆ |
| 20 | CL-OP55H | claude-opus-5.5-high | Anthropic | 717 | 1504 | — | 0.770 | 0.589 | 0.778 | — | — | — | ★☆☆ |
| 21 | CL-OP47H | Claude Opus 4.7 High | Anthropic | 708 | 1501 | — | 0.755 | 0.589 | 0.778 | — | — | — | ★☆☆ |
| 22 | GEM-38FH | Gemini 3.8 Flash High | 689 | 1495 | 1583 | 0.739 | 0.554 | 0.778 | — | — | — | ★☆☆ | |
| 23 | GPT-56S | GPT-5.6 Sol | OpenAI | 581 | 1484 | — | 0.670 | 0.305 | 0.803 | $2.00 | $10.00 | 1M | ★★☆ |
| 24 | CL-OP47 | Claude Opus 4.7 | Anthropic | 546 | 1494 | — | 0.720 | 0.248 | 0.502 | $5.00 | $25.00 | 1M | ★★☆ |
| 25 | INKL | Inkling | Thinking Machines | 531 | — | — | 0.379 | 0.613 | 0.923 | $1.87 | $4.68 | 256K | ★☆☆ |
| 26 | GEM-3P | Gemini 3 Pro | 495 | 1485 | — | 0.675 | 0.028 | 0.766 | $2.00 | $12.00 | 66K | ★★☆ | |
| 27 | CL-OP46 | Claude Opus 4.6 | Anthropic | 491 | 1497 | — | 0.735 | 0.037 | 0.502 | $5.00 | $25.00 | 1M | ★★☆ |
| 28 | GEM-31P | Gemini 3.1 Pro Preview | 487 | 1487 | — | 0.544 | 0.241 | 0.766 | $2.00 | $12.00 | 1M | ★★★ | |
| 29 | CL-OP48 | Claude Opus 4.8 | Anthropic | 484 | 1494 | — | 0.720 | 0.042 | 0.502 | $5.00 | $25.00 | 1M | ★☆☆ |
| 30 | GPT-6L | GPT-6 Luna | OpenAI | 436 | 1440 | — | 0.450 | 0.130 | 0.994 | $0.10 | $0.50 | 128K | ★★☆ |
Methodology
Composite score = round((tape × 0.55 + fundamentals × 0.30 + cost_index × 0.15) × 1000)
- Tape (55%) — Fast-moving market signal built from Chatbot Arena text/code plus Artificial Analysis. Arena text carries 65% of the arena blend, code 35%, then that arena composite gets blended 60/40 with the AA Intelligence Index.
- Fundamentals (30%) — Slower benchmark layer built from Terminal Bench, LiveBench, a demoted legacy coding/Tau fallback, and a light AA capability proxy. Trust penalties and freshness decay hit each source before the blend.
- Cost Idx (15%) — Cost efficiency 0–1. 1 = cheapest, 0 = most expensive. Blended cost = input 25% + output 75%, min-max normalized across all tracked models. Sourced from OpenRouter.
Column glossary
- Score — Composite score 0–1000. Higher is better.
- Text ELO — Raw Arena text leaderboard ELO.
- Code ELO — Raw Arena code leaderboard ELO.
- Tape — Fast, community-sensitive score from Arena plus Artificial Analysis.
- Fund — Slower benchmark score from Terminal Bench, LiveBench, and lower-weight fallback fundamentals.
- Cost Idx — Cost efficiency index 0–1. 1 = best value in the current tracked set.
- $/M in — Input cost per 1 million tokens (USD), via OpenRouter.
- $/M out — Output cost per 1 million tokens (USD), via OpenRouter.
- Ctx — Context window size (max tokens the model can process in one request).
- ★★★ — No full-component fallback used. ★★☆ — 1 component imputed. ★☆☆ — 2 components imputed.
SFX-10 is the arithmetic mean composite of the top 10 models. Constituents rotate as scores change. Refreshed every 6 hours.