2026年11月のベストAIモデル:最新ランキングとリリース年表
2026年11月スナップショット:Intelligence Index v5.0とCoding Agent Index v2.0の順位、GPT-6・Claude 5.5・Gemini 4・Grok 4.7のAPI価格、6月から11月までの日付付きリリース年表。
Best AI Models in November 2026: Current Leaderboard + Release Timeline
Most “best AI models” lists you’ll find are frozen in July. The frontier moved three times since then: OpenAI shipped GPT-6 Astra and the GPT-6.1 line, Anthropic shipped Opus 5.5, and Google made Gemini 4 Argon generally available. This page is the November 2026 snapshot — leaderboard, pricing, and a dated release timeline from June onward — assembled from the Artificial Analysis indexes as reported on November 3, 2026 and vendor documentation, cross-checked against our own model reviews on November 20, 2026.
The November Leaderboard (Intelligence Index v5.0)
Artificial Analysis’ Intelligence Index combines knowledge work, coding, science and general knowledge. Scores are index points, not percentages; effort/fallback configurations are kept as published:
| Model (configuration) | Index v5.0 points | Access note |
|---|---|---|
| Gemini 4 Carbon (high) | 59 | Restricted rollout — cyber defenders / trusted partners |
| Claude Opus 5.5 (max, fallback) | 58 | Public API |
| GPT-6 Astra (max) | 57 | Public API + ChatGPT |
| GPT-6.1 Sol (max) | 55 | Public API |
| Claude Sonnet 5.5 (max, fallback) | 54 | Public API |
| Gemini 4 Argon (high) | 50 | GA since Oct 13 |
| Claude Fable 5.1 (max, fallback) | 49 | Public API |
| Grok 4.7 (xhigh) | 35 | xAI API |
Two caveats worth keeping: the top score (Carbon) belongs to a model most readers cannot access yet, and the index version changed in October (v1.5 → v2.0 for the coding index) — a lower number than a July list does not by itself mean a model got worse.
Coding Agents: Score Is Only Half the Story
The Coding Agent Index v2.0 (Nov 3 snapshot) scores complete agent setups — model plus tooling — and publishes cost and time per task:
| Agent setup | Index points | API cost / task | Active time / task |
|---|---|---|---|
| Claude Code + Opus 5.5 (max, fallback) | 62 | $11.42 | 1.0 h |
| Antigravity CLI + Gemini 4 Carbon (high) | 59 | $7.74 | 29.8 min |
| Claude Code + Sonnet 5.5 (max, fallback) | 59 | $13.30 | 1.3 h |
| Codex + GPT-6 Astra (max) | 55 | $7.04 | 24.8 min |
| Codex + GPT-6.1 Sol (max) | 54 | $1.50 | 23.7 min |
| Grok Build + Grok 4.7 (xhigh) | 47 | $9.48 | 35.8 min |
The standout row is Codex + GPT-6.1 Sol: one point behind Astra at roughly a fifth of the task cost. These are benchmark averages — measure on your own repositories before budgeting.
API Pricing Snapshot (per 1M tokens, standard uncached)
| Model | Input | Output | Fine print |
|---|---|---|---|
| GPT-6 Astra | $10 | $50 | ≤272K input; above that, rates rise |
| GPT-6.1 Sol | $2 | $10 | Same short-context scope |
| GPT-6.1 Terra | $1 | $5 | Visual / computer-use positioning |
| Claude Fable 5.1 | $10 | $50 | Cached input $0.25 (per our review) |
| Claude Opus 5.5 | $4 | $20 | Base API rates |
| Claude Sonnet 5.5 | $2 | $10 | Base API rates |
| Gemini 4 Argon | $2 | $10 | Introductory through Nov 30; $4/$20 from Dec 1 |
| Grok 4.7 | $2 | $6 | <200K prompt; $4/$12 at 200K+ |
Sticker prices exclude thinking tokens, cache writes, tools and repeated agent turns. The useful number is the full invoice for an accepted result.
Release Timeline: June → November 2026
- June 9 — Anthropic ships Claude Fable 5, the safety-tiered Mythos-class release ($10/$50, Opus 4.8 fallback).
- July 9 — GPT-5.6 reaches GA in Sol / Terra / Luna tiers after a 12-day government review; July 23 GPT-5.4 retires; July 30 the price cut lands (Sol $5/$30, Terra $2/$12, Luna $0.20/$1.20). If you searched “new AI models July 2026”, this was the month’s story.
- September 3 — GPT-6 Astra: reasoning-first flagship, first “Critical” cyber threshold, 1.1M-token input context.
- September 22 — Claude Opus 5.5 launches; the lineup now runs Opus / Fable / Sonnet / Haiku 5.5.
- September 29 — GPT-6.1 Sol: near-Astra coding at $2/$10.
- October 12–13 — Meta Muse Wonder arrives; Gemini 4 Argon becomes generally available.
- October 26 — GPT-6.1 Terra ($1/$5) for visual / computer-use work.
- October 28 — Gemini 4 Carbon announced with restricted rollout; tops the public index but is not broadly accessible.
Video Models: The Other Half of the Stack
For creators, the video side moved too: Kuaishou shipped Kling 3.0 (Feb 5, native 4K), Alibaba’s Wan 2.7 added Thinking Mode and native audio (April), and MiniMax’s H3 Max brought stereo-audio video up to 15 seconds. On ZNIX the live line is Kling 2.6, Seedance 2.0, Hailuo 2.3, H3 Max and Vidu — see the model hub and our Kling vs Runway vs Sora 2 vs Luma comparison for the access-and-pricing breakdown. Sora 2 remains ChatGPT-only with no API; the original Sora shut down April 26, 2026.
Which Should You Actually Use?
- Daily coding: start with Claude Opus 5.5 or GPT-6.1 Sol; add Sonnet 5.5 when token price matters.
- Hard reasoning / research: test GPT-6 Astra or Fable 5.1 against that baseline; keep the premium model only if it improves accepted results.
- Multimodal research: Gemini 4 Argon (public) while Carbon’s rollout stays restricted.
- Video production: credit-based models with real APIs — see the comparison above.
Sources: Artificial Analysis Intelligence Index v5.0 and Coding Agent Index v2.0 snapshots (Nov 3, 2026) and vendor pricing pages, as compiled by divkix.me/blog/ai-models-compared-2026 (secondary source, checked 2026-11-20); release dates and tier details cross-checked against our Astra review, Fable 5 explainer and GPT-5.6 explainer. Index points are not percentages; configurations and access limits are kept as published. This is a dated snapshot, not a live leaderboard.
よくある質問
What is the best AI model in November 2026?
What new AI models were released in July 2026?
Why do November scores look lower than July lists?
Are these prices and scores live?
さらに読む
ZNIX 編集チームは一次資料へのリンク、日付付きの提供状況確認、明示した例を用いて AI モデルを解説します。資料の要約と実測テストを区別しています。