Get the TERNZ AI weekly newsletter: the week's top AI travel tools and AI news, by email.

Intelligence: The Most Powerful AI Models, Ranked

The most powerful AI models on Earth — closed and open

FRONTIER POWER RANKINGS

Closed Frontier Models

Scored on Edge + Alpha: raw intelligence, real-world usage and profitability.

Closed sourceScore = Edge + Alpha (0–100)
1Claude Fable 5.2Anthropic · USA · Closed
99
Released November 9: Anthropic’s new most powerful generally available model. In Anthropic’s launch comparison, Fable 5.2 or its cyber twin Mythos 5.2 leads 13 of 15 benchmark lines, including FrontierCode 2.0 (49.1% vs. 42.8% for GPT-6 Astra) and FrontierSWE v2 (69.9% vs. 65.5%) — company-reported numbers. API prices drop 20% to $8 and $40 per million tokens; Mythos 5.2 stays limited to vetted cyber and life-sciences teams.
2GPT-6 AstraOpenAI · USA · Closed
98
Released September 3: OpenAI's most capable model yet and a new state of the art for computer and browser use. On OpenAI's own numbers it posts 97.6% on FrontierMath Tier 4, 74.1% on DeepSWE and 100% on ExploitBench. Controversial, too — its "opaque recurrence" reasoning makes chain-of-thought monitoring harder.
3Claude Opus 5.5Anthropic · USA · Closed
97
Released September 22 and called by Anthropic “the strongest-performing model we’ve tested to date,” outpacing the larger Fable model on many benchmarks. Output tokens drop to $20 per million from $25, and the model runs faster on less compute. The first Anthropic release since CEO Dario Amodei embraced pacing the frontier — alignment progress first, capability second.
4Gemini 4 ArgonGoogle · USA · Closed
95
Announced September 30 and broadly available since October 13, with introductory pricing of $2 and $10 per million tokens through November 30 before it doubles. It tops the Vals Index at 68.9%, yet ties GPT-6 Astra at 53 on the Artificial Analysis index, behind Claude Opus 5.5 (58) — and Opus 5.5 still wins on Terminal-Bench.

Open-Source Models from Around the World

Open-weight models anyone can download, run and build on.

Open sourceScore = Edge + Alpha (0–100)
1Kimi K3Moonshot AI · China · Open
90
Now fully open: Moonshot released the weights of the largest open model ever — 2.8T parameters, 1M-token context, Opus-class coding and agents. A landmark moment for open AI.
2DeepSeek V4.1 FlashDeepSeek · China · Open
89
The cost disruptor, now the volume leader: V4.1 Flash shipped earlier this month and sits at number one on OpenRouter’s leaderboard after a 172% jump in usage in a single week. Real-world adoption, not just a benchmark line — pure alpha.
3Qwen 3.7 MaxAlibaba · China · Open
86
The most-run open model on Earth — permissive license, every size, the default local brain for builders worldwide.

Top U.S. Open-Source Models

Meta’s flagship open model, plus NVIDIA — tracked every week.

Meta · USA · Open weights
Muse Spark Open
Announced October 15, 2026: an 816B-parameter mixture-of-experts model (about 42B active) with a 1M-token context that takes in text, images, video and audio. Meta calls it the most capable U.S. open-weight model to date and reports 39 on the Artificial Analysis Intelligence Index (not yet independently verified), still below the leading Chinese open models. Custom Muse license.
Source: VentureBeat, Oct 15, 2026 → ai.meta.com →
NVIDIA · USA · Open weights
Nemotron 3 Ultra
NVIDIA’s flagship open model, released June 2026: a 550B-parameter hybrid Mamba-Transformer mixture-of-experts model (55B active), with weights, data and training recipes published under the permissive OpenMDW license. NVIDIA reports about 5x the throughput of comparable open models. It scores 23 on the Artificial Analysis Intelligence Index (per VentureBeat), behind Muse Spark Open.
New this week: Nemotron 3.5 Bolt E4B (about 4GB, October 23), built on Gemma 4 E4B for on-device voice assistants and robotics. NVIDIA reports about 40% higher average quality than Nemotron 3 Nano 4B on conversational-agent benchmarks (company-reported). Earlier: Nemotron 3.5 Super VL (120B, October 15) scores about 26 on the Artificial Analysis index, and Nemotron 3.5 Bolt (74B, 9B active) fits in 24GB for local agents.
Source: NVIDIA, June 2026 → Source: NVIDIA, Oct 15, 2026 → Source: VentureBeat, Oct 15, 2026 → Source: NVIDIA Research, Oct 23, 2026 → research.nvidia.com/labs/nemotron →
Also watching: Reflection AI’s Beam (announced October 5; weights pending) and Thinking Machines Lab’s Inkling (26 on the index, per VentureBeat).
Composite of intelligence, adoption & profitability · October 2026 · for orientation, not investment advice.