Autonomous multi-source actuarial benchmark evaluating 50 frontier and mid-tier cognitive architectures with hard contamination-resistant private tasks (29%), coding & SWE (18%), reasoning (14%), agentic computer use (13%), openness rewards (5%), permanent 90-day velocity bonuses, and subtractive anti-hype penalties.
| Rank & Model | Developer | License | AKI Score | Confidence | Reasoning | Agent Code | Velocity | Pricing / 1M | Why Ranked Here |
|---|---|---|---|---|---|---|---|---|---|
| #1 Gemini 4 Argon High | Proprietary | 93.2 | 88% (1.8) | 98.6 | 94.2 | +2.88 | $2.00 / $10.00 | Why #1: Strong early independent signals + Arena + high velocity — Top drivers: Hard Capability 93rd percentile + Agentic 94th percentile — Rate of improvement +4.8 last 90 days → velocity bonus +2.88 — Social velocity penalty 0 — Pricing $2/$10 excellent value — Forward-looking #1. | |
| #2 GPT-6 Astra Max | OpenAI | Proprietary | 92 | 93% (1.1) | 99.4 | 81.4 | +0.72 | $10.00 / $50.00 | Why #2: Extremely broad independent evidence — Top drivers: Hard Capability 96th percentile + Reasoning 95th percentile — Social velocity penalty -5 applied — Pricing $10/$50 — Highest multi-source consensus — Slightly reduced historical advantage Hard 32%→29%. |
| #3 Claude Opus 5.5 Max/Thinking | Anthropic | Proprietary | 90.1 | 93% (1.1) | 99.1 | 94 | +0.50 | $4.00 / $20.00 | Why #3: Deep independent evidence base — Top drivers: Coding & SWE 94th percentile + Hard 92nd percentile — Social velocity penalty -7.5 applied — Still elite hype filtered — Pricing $4/$20 — Confidence 95%→93% reduced historical advantage. |
| #4 Llama Spark 1.3 Max | Meta | Open Weights | 88.4 | 90% (1.5) | 94.5 | 91 | +1.86 | $1.25 / $4.25 | Why #4: Solid agentic + multimodal + velocity — Top drivers: Agentic 91st percentile + Openness 95th percentile open-weight — Velocity +1.86 — Agentic 11%→13% +1.0 + openness 3%→5% +1.0 — Pricing ~$1.25/$4.25 good price — Rising. |
| #5 Claude Sonnet 5.5 Max | Anthropic | Proprietary | 88.2 | 91% (1.4) | 96 | 90 | +0.40 | $2.00 / $10.00 | Why #5: Excellent breadth + value — Top drivers: Coding 90th percentile + Efficiency 89th percentile — Velocity +0.4 — Best pure value top tier — Outstanding coding/agent + value less hype than Opus/Fable. |
| #6 Grok 4.7 xhigh | xAI | Proprietary | 88 | 88% (1.8) | 95 | 93 | +3.00 | $2.00 / $6.00 | Why #6: Clear agentic & coding progress + velocity — Top drivers: Agentic 93rd percentile + Coding 87th percentile — Velocity +3 cap fastest improvement +5.2 last 90 days — Pricing $2/$6 — X-native tools. |
| #7 Qwen 3.8 Max | Alibaba | Open Weights | 86.1 | 90% (1.5) | 94 | 89 | +1.68 | $0.40 / $0.80 | Why #7: Strong coding + multilingual + open — Top drivers: Coding 89th percentile + Openness 94th percentile open-weight — Velocity +1.68 — Openness +1.0 — Leading open-weight. |
| #8 DeepSeek V4 Pro Max | DeepSeek | Open Weights | 85.2 | 90% (1.5) | 97 | 86 | +1.44 | $0.66 / $1.98 | Why #8: Strong open-weight reasoning + value + velocity — Top drivers: Efficiency 92nd percentile best pure value + Openness 95th percentile — Velocity +1.44 — Extreme cost efficiency + open. |
| #9 GLM-5.3 Max | Z.AI | Open Weights | 85 | 89% (1.6) | 93 | 85 | +1.20 | $1.40 / $4.40 | Why #9: Competitive reasoning/coding + open — Top drivers: Reasoning 87th percentile + Openness 93rd percentile — Openness +1.0 — Good value. |
| #10 Kimi K3 Max | Moonshot | Proprietary | 84.8 | 90% (1.5) | 94 | 86 | +0.60 | $2.70 / $13.50 | Why #10: Exceptional long-context + velocity — Top drivers: Reasoning 88th percentile + Agentic 86th percentile long-context — Velocity +0.6 — Solid strong reasoning & long context. |
| #11 GPT-6.1 Sol Max | OpenAI | Proprietary | 84.5 | 87% (1.9) | 93 | 84 | +0.40 | $2.00 / $10.00 | Why #11: Near-Astra intelligence at low cost — Top drivers: Efficiency 91st percentile + Hard 85th percentile — Near-Astra at ~1/5 cost — Efficiency king — Was 88.9 #5 now #11 because hard reduction hits efficiency-focused model. |
| #12 Claude Fable 5.1 Max | Anthropic | Proprietary | 84.3 | 91% (1.4) | 99.6 | 88 | +0.20 | $10.00 / $50.00 | Why #12: Strong coding & long-horizon — Top drivers: Coding 88th percentile + Agentic 87th percentile long-horizon — Strong coding & long-horizon — Still top-tier after penalty expensive + high social noise. |
| #13 MiMo-V2.6-Pro | Xiaomi | Open Weights | 83.2 | 87% (1.9) | 89 | 84 | +1.20 | $0.40 / $0.80 | Why #13: Leading open-weight contender + velocity — Top drivers: Openness 96th percentile open-weight + Efficiency 90th percentile $0.40/$0.80 best value — Velocity +1.2 — Best open-weight overall. |
| #14 Claude Opus 5 | Anthropic | Proprietary | 80.9 | 89% (1.6) | 92 | 83 | +0.10 | $5.00 / $25.00 | Why #14: Deep historical evidence — Top drivers: Hard 84th percentile + Coding 83rd percentile — Previous-gen but still competitive — Deep historical evidence. |
| #15 GPT-6 Sol Max | OpenAI | Proprietary | 80.2 | 85% (2.2) | 90 | 82 | +0.10 | $2.00 / $10.00 | Why #15: Balanced everyday frontier — Top drivers: Efficiency 86th percentile + Reasoning 82nd percentile — Reliable — Balanced frontier everyday model. |
| #16 DeepSeek V4.1 Flash Max | DeepSeek | Open Weights | 80.5 | 89% (1.6) | 88 | 81 | +1.10 | $0.20 / $0.80 | Why #16: Extreme efficiency + open — Top drivers: Efficiency 94th percentile budget king + Openness 94th percentile — Openness + velocity — Budget king cheapest high-performer. |
| #17 Gemini 3.8 Flash | Proprietary | 78.9 | 88% (1.8) | 87 | 82 | +0.40 | $0.75 / $3.75 | Why #17: Speed + price — Top drivers: Efficiency 88th percentile + Agentic 82nd percentile — High-volume workhorse. | |
| #18 Step 5 Preview | StepFun | Proprietary | 78.8 | 84% (2.4) | 88 | 78 | +1.20 | $0.50 / $2.00 | Why #18: Emerging frontier — Top drivers: Reasoning 82nd percentile + Hard 80th percentile — Velocity +1.2 — Emerging frontier. |
| #19 Grok 4.6 | xAI | Proprietary | 77.3 | 85% (2.2) | 86 | 81 | +0.30 | $2.00 / $6.00 | Why #19: Previous generation — Top drivers: Coding 81st percentile + Agentic 80th percentile — Previous Grok generation. |
| #20 Claude Fable 5 | Anthropic | Proprietary | 75.8 | 89% (1.6) | 89 | 82 | +0.10 | $10.00 / $50.00 | Why #20: Deep evidence base — Top drivers: Coding 82nd percentile + Reasoning 81st percentile — Deep evidence base. |
| #21 Qwen 3.8 27B | Alibaba | Open Weights | 77 | 87% (1.9) | 85 | 84 | +1.20 | $0.20 / $0.60 | Why #21: Strong open model + open — Top drivers: Openness 93rd percentile + Coding 84th percentile — Strong open model. |
| #22 GLM-5.3 Flash | Z.AI | Open Weights | 76.2 | 86% (2.1) | 84 | 80 | +1.10 | $0.15 / $0.50 | Why #22: High efficiency open + open — Top drivers: Efficiency 92nd percentile extreme value + Openness 92nd percentile — Extreme value. |
| #23 DeepSeek V4 Flash | DeepSeek | Open Weights | 75.8 | 85% (2.2) | 83 | 79 | +1.00 | $0.14 / $0.28 | Why #23: Budget high-performer + open — Top drivers: Efficiency 93rd percentile + Openness 93rd percentile — Budget high-performer. |
| #24 GPT-6 Luna | OpenAI | Proprietary | 72.8 | 80% (3) | 80 | 76 | +0.20 | $0.10 / $0.50 | Why #24: Ultra-cheap tier — Top drivers: Efficiency 90th percentile ultra-cheap + Freshness 80th percentile — High-volume only not intelligence-focused. |
| #25 MiniMax M3 | MiniMax | Open Weights | 74 | 81% (2.8) | 82 | 75 | +1.10 | $0.30 / $1.20 | Why #25: Growing evidence + open — Top drivers: Openness 88th percentile + Reasoning 78th percentile — Growing evidence — Emerging. |
| #26 GPT-6 Mini | OpenAI | Proprietary | 73.5 | 82% (2.7) | 81 | 75 | +0.80 | $0.20 / $1.00 | Why #26: New entry #26 — GPT-6 Mini — AKI 73.5 · Conf 82% — Near-GPT-6 Sol intelligence at mini cost — Top drivers: Efficiency 91st percentile + Reasoning 79th percentile — Freshness boost +8% new high-quality independent evidence <14 days. |
| #27 Claude Haiku 5.5 | Anthropic | Proprietary | 73 | 84% (2.4) | 80 | 80 | +0.60 | $0.50 / $2.50 | Why #27: New entry #27 — Claude Haiku 5.5 — AKI 73.0 · Conf 84% — Fast + cheap Claude — Top drivers: Efficiency 89th percentile + Coding 80th percentile — Social velocity penalty -3 — Pricing $0.50/$2.50 — Emerging. |
| #28 Gemini 4 Flash High | Proprietary | 72.8 | 83% (2.5) | 81 | 83 | +0.90 | $0.50 / $2.00 | Why #28: New entry #28 — Gemini 4 Flash High — AKI 72.8 · Conf 83% — Speed + Argon lineage — Top drivers: Efficiency 90th percentile + Agentic 83rd percentile — Freshness boost +10% — Pricing $0.50/$2.00 — Emerging. | |
| #29 Kimi K4 Max | Moonshot | Proprietary | 71.8 | 82% (2.7) | 84 | 77 | +0.70 | $2.50 / $12.00 | Why #29: New entry #29 — Kimi K4 Max — AKI 71.8 · Conf 82% — Exceptional long-context 1M+ — Top drivers: Reasoning 84th percentile + Openness 90th percentile — Freshness boost +8% new high-quality independent evidence <14 days. |
| #30 DeepSeek V5 Pro | DeepSeek | Open Weights | 71.5 | 83% (2.5) | 82 | 82 | +1.50 | $0.70 / $2.10 | Why #30: New entry #30 — DeepSeek V5 Pro — AKI 71.5 · Conf 83% — Next-gen open-weight — Top drivers: Openness 95th percentile + Coding 82nd percentile — Velocity +1.5 — Pricing ~$0.70/$2.10 — Emerging. |
| #31 Qwen 3.9 Max | Alibaba | Open Weights | 70.8 | 84% (2.4) | 82 | 85 | +1.20 | $2.20 / $6.50 | Why #31: New entry #31 — Qwen 3.9 Max — AKI 70.8 · Conf 84% — Coding + multilingual + open — Top drivers: Coding 85th percentile + Openness 94th percentile — Velocity +1.2 — Pricing $2.20/$6.50 — Emerging. |
| #32 GLM-6 Preview | Z.AI | Open Weights | 70.2 | 81% (2.8) | 83 | 78 | +1.00 | $1.50 / $4.80 | Why #32: New entry #32 — GLM-6 Preview — AKI 70.2 · Conf 81% — Next-gen reasoning + open — Top drivers: Reasoning 83rd percentile + Openness 92nd percentile — Freshness boost +12% — Pricing ~$1.50/$4.80 — Emerging. |
| #33 MiMo-V3 Preview | Xiaomi | Open Weights | 69.8 | 80% (3) | 79 | 77 | +1.00 | $0.50 / $1.00 | Why #33: New entry #33 — MiMo-V3 Preview — AKI 69.8 · Conf 80% — Next open-weight contender — Top drivers: Openness 95th percentile + Efficiency 88th percentile — Velocity +1.0 — Pricing ~$0.50/$1.00 — Emerging. |
| #34 Llama Spark 1.2 | Meta | Open Weights | 69.5 | 86% (2.1) | 80 | 82 | +0.10 | $1.00 / $3.50 | Why #34: Previous Llama — Top drivers: Agentic 82nd percentile + Openness 90th percentile — Previous-gen but still solid agentic + multimodal — Deep historical evidence. |
| #35 Grok 4.5 | xAI | Proprietary | 69 | 83% (2.5) | 79 | 79 | +0.10 | $2.00 / $6.00 | Why #35: Previous Grok — Top drivers: Coding 79th percentile + Agentic 78th percentile — Previous generation. |
| #36 Claude Opus 4.8 | Anthropic | Proprietary | 68.5 | 88% (1.8) | 84 | 77 | +0.05 | $5.00 / $25.00 | Why #36: Legacy but still used — Top drivers: Hard 78th percentile + Coding 77th percentile — Deep historical evidence — Legacy. |
| #37 GPT-4o / GPT-5 Legacy | OpenAI | Proprietary | 68 | 90% (1.5) | 82 | 75 | +0.00 | $2.50 / $10.00 | Why #37: Legacy frontier — Top drivers: Hard 77th percentile + Reasoning 76th percentile — Deep historical evidence — Legacy but still used. |
| #38 Gemini 2.5 Pro | Proprietary | 67.5 | 89% (1.6) | 81 | 74 | +0.00 | $1.25 / $5.00 | Why #38: Legacy Gemini — Top drivers: Reasoning 77th percentile + Efficiency 76th percentile — Deep historical evidence. | |
| #39 Kimi K2.x | Moonshot | Proprietary | 67 | 82% (2.7) | 80 | 75 | +0.00 | $1.20 / $4.80 | Why #39: Previous Kimi — Top drivers: Reasoning 76th percentile + Agentic 75th percentile — Previous generation. |
| #40 Qwen 3.5 Max | Alibaba | Open Weights | 66.5 | 85% (2.2) | 78 | 77 | +0.00 | $0.80 / $2.40 | Why #40: Previous Qwen — Top drivers: Coding 77th percentile + Openness 90th percentile — Previous-gen. |
| #41 DeepSeek V3 | DeepSeek | Open Weights | 66 | 87% (1.9) | 79 | 74 | +0.00 | $0.14 / $0.28 | Why #41: Previous DeepSeek — Top drivers: Efficiency 87th percentile + Openness 92nd percentile — Previous-gen — Extreme cost efficiency. |
| #42 Mistral Large 3 | Mistral | Open Weights | 65.5 | 83% (2.5) | 77 | 73 | +0.00 | $2.00 / $6.00 | Why #42: European open-weight — Top drivers: Openness 91st percentile + Reasoning 74th percentile — Competitive open-weight. |
| #43 Command R+ / Cohere | Cohere | Proprietary | 65 | 82% (2.7) | 76 | 74 | +0.00 | $3.00 / $15.00 | Why #43: RAG + enterprise — Top drivers: Science & Knowledge 76th percentile + Agentic 74th percentile — Enterprise-focused. |
| #44 Claude Sonnet 4.5 | Anthropic | Proprietary | 64.5 | 87% (1.9) | 78 | 75 | +0.00 | $3.00 / $15.00 | Why #44: Previous Sonnet — Top drivers: Coding 75th percentile + Efficiency 73rd percentile — Previous-gen. |
| #45 GPT-6 Nano | OpenAI | Proprietary | 64 | 80% (3) | 73 | 71 | +0.50 | $0.05 / $0.20 | Why #45: New entry #45 — GPT-6 Nano — AKI 64.0 · Conf 80% — Ultra-cheap nano — Top drivers: Efficiency 92nd percentile $0.05/$0.20 — Freshness boost +10% — High-volume only not intelligence-focused. |
| #46 Gemini 3.5 Flash | Proprietary | 63.5 | 86% (2.1) | 75 | 72 | +0.00 | $0.50 / $2.00 | Why #46: Previous Flash — Top drivers: Efficiency 85th percentile + Speed — High-volume workhorse — Previous-gen. | |
| #47 Phi-4 / Microsoft | Microsoft | Open Weights | 63 | 84% (2.4) | 76 | 70 | +0.00 | $0.10 / $0.30 | Why #47: Small open model — Top drivers: Openness 90th percentile + Efficiency 84th percentile — Small but efficient — Open. |
| #48 Llama 4 Scout | Meta | Open Weights | 62.5 | 85% (2.2) | 74 | 76 | +0.00 | $0.80 / $3.00 | Why #48: Previous Llama scout — Top drivers: Agentic 76th percentile + Openness 89th percentile — Previous-gen. |
| #49 Gemma 3 27B | Open Weights | 62 | 83% (2.5) | 73 | 70 | +0.00 | $0.15 / $0.40 | Why #49: Small open Google — Top drivers: Openness 91st percentile + Efficiency 82nd percentile — Small open-weight — Good value. | |
| #50 MiniMax M2 | MiniMax | Open Weights | 61.5 | 79% (3.1) | 71 | 68 | +0.00 | $0.20 / $0.80 | Why #50: Previous MiniMax — Top drivers: Openness 87th percentile + Reasoning 71st percentile — Growing evidence — Previous-gen — 15 slots reserved for next-gen entries. |
Under the AKI Grounded Intelligence Engine™ v2.2 Anti-Softness release, Google Gemini 4 Argon High ranks #1 with an AKI score of 93.2/100 (Confidence 88%, uncertainty ±1.8, velocity bonus +2.88). It is followed by OpenAI GPT-6 Astra Max (Rank #2, AKI 92.0, Confidence 93%) and Anthropic Claude Opus 5.5 Max (Rank #3, AKI 90.1, Confidence 93% after social velocity filtering). (Reconciled: October 3, 2026)
Meta's Llama Spark 1.3 Max ranks as the #1 open-weights foundation model (Rank #4 overall, AKI Score 88.4, Confidence 90%, ±1.5), followed by Alibaba Qwen 3.8 Max (Rank #7, AKI 86.1, Confidence 90%) and DeepSeek V4 Pro Max (Rank #8, AKI 85.2, Confidence 90%). Open-weights architectures capture 44% (22 of 50) of the master ledger. (Reconciled: October 3, 2026)
DeepSeek V4.1 Flash Max achieves the highest efficiency score at $0.20 input / $0.80 output per 1M tokens (Rank #16, AKI 80.5, 94th percentile efficiency). In the top 10, Alibaba Qwen 3.8 Max ($0.40 in / $0.80 out, Rank #7) and Claude Sonnet 5.5 Max ($2.00 in / $10.00 out, Rank #5) deliver the highest value density.
The v2.2 formula requires multi-source consensus across hard benchmarks: 29% Hard Capability (AA Private Set, HLE, FrontierMath), 18% Coding & SWE, 14% Reasoning & Formal Math, 13% Agentic Computer Use (OSWorld 2.0), 10% Science & Knowledge, 7% Efficiency, 5% Openness, 3% Freshness, and 1% Preference Diversity. Subtractive social velocity penalties deduct score if social buzz outpaces independent evaluation, and a 90-day velocity bonus rewards rapid real improvement.