星図を広げる
おすすめ 架空モデル ホシゾラ-9
設定 effort high / context 長め
仮。広い範囲を一度に見るときのサンプル。
AIチーム用モデルカタログ。正本は catalog/models.yaml
Quest の種類ごとのおすすめ。正本は catalog/erabi.md
仮サンプル。本番の推薦文ではない。種類もモデルも架空。ビルド確認用。
おすすめ 架空モデル ホシゾラ-9
設定 effort high / context 長め
仮。広い範囲を一度に見るときのサンプル。
おすすめ 架空モデル コガタ-2
設定 effort medium / 速め
仮。短い往復で手を動かすときのサンプル。
おすすめ 架空モデル メガネ-1
設定 effort xhigh / 遅めでもよい
仮。差分を厳しく読むときのサンプル。
| # | モデル | 提供元 | 状態 | コンテキスト | Effort既定 | コスト | 速度 | T3 | 確認日 |
|---|---|---|---|---|---|---|---|---|---|
| 1 |
claude-opus-5 |
anthropic | active | 1,000,000 | high | 高 | 普通 | カニ様（未確認） | 2026-09-28 |
| 2 |
claude-opus-5-5 |
anthropic | active | 1,000,000 | medium | 高 | 普通 | カニ様（未確認） | 2026-09-28 |
| 3 |
claude-sonnet-5 |
anthropic | active | 1,000,000 | high | 中 | 速め | カニ様: Claude Sonnet 5 | 2026-09-24 |
| 4 |
claude-fable-5-1 |
anthropic | active | 1,000,000 | high | 特高 | 遅め | カニ様（未確認） | 2026-09-24 |
| 5 |
gpt-5.6-sol |
openai | active | 1,050,000 | medium | 高 | 速め | エビ様（未確認） | 2026-09-24 |
| 6 |
gpt-5.6-terra |
openai | active | 1,050,000 | medium | 中 | 速め | エビ様（未確認） | 2026-09-28 |
| 7 |
gpt-5.6-luna |
openai | active | 1,050,000 | medium | 低 | 速め | エビ様（未確認） | 2026-09-28 |
| 8 |
gpt-6-astra |
openai | active | 1,050,000 | unknown | 特高 | 速め | エビ様: GPT-6-Astra | 2026-09-28 |
| 9 |
gpt-6-sol |
openai | active | 1,050,000 | medium | 中 | 速め | エビ様（未確認） | 2026-09-28 |
| 10 |
gpt-6-luna |
openai | active | 1,050,000 | medium | 低 | 速め | エビ様（未確認） | 2026-09-28 |
| 11 |
gemini-3.8-flash |
active | 1,048,576 | medium | 低 | unknown | カイ様: Gemini 3.8 Flash (High), Gemini 3.8 Flash (Medium), Gemini 3.8 Flash (Low) | 2026-09-24 | |
| 12 |
gemini-3.1-pro-preview |
preview | 1,048,576 | high | 中 | unknown | カイ様（未確認） | 2026-09-24 | |
| 13 |
grok-4.6 |
xai | active | 500,000 | high | 中 | unknown | かーくん: Cursor Grok 4.6 / ウニ様: Grok Build | 2026-09-24 |
| 14 |
grok-4.7 |
xai | active | 500,000 | high | 中 | unknown | かーくん（未確認） / ウニ様（未確認） | 2026-09-24 |
claude-opus-5 · claude-opus-5
Complex agentic coding and enterprise work; strong verification/iteration on long coding tasks (Anthropic models overview + Opus 5 launch)
Active (legacy); Anthropic recommends migrating to Opus 5.5. Adaptive thinking; API effort default high. Pricing $5/$25 per MTok. Comparative latency Moderate → speed_class 普通. Output $25 → cost_class 高. Claude Code Anthropic API opus alias resolved to Opus 5 through v2.1.279; from v2.1.280+ opus (and account defaults on Anthropic API) → Opus 5.5. T3 labels/menus unknown (eri 2026-09-17).
claude-opus-5-5 · claude-opus-5-5
Long-running agentic coding and knowledge work; first Claude 5.5 Opus; performs near Fable 5.1 on most work at lower cost (Anthropic Opus 5.5 announcement + overview)
Adaptive thinking always on. API effort default medium (unlike Opus 5 high). All five effort levels. Pricing $4/$20 per MTok. Comparative latency Moderate → speed_class 普通. Output $20 → cost_class 高. Released 2026-09-22. Status Active (latest). Max output 128K. Claude Code Anthropic API: opus alias + Pro/Max/Team/Enterprise/API defaults → Opus 5.5 (v2.1.280+; was Opus 5 before). T3 labels/menus unknown (not yet confirmed by eri).
claude-sonnet-5 · claude-sonnet-5
Best speed/intelligence mix; agentic coding, tool use, and knowledge work at Sonnet price (Anthropic models overview + Sonnet 5 launch)
Adaptive thinking; API effort default high. Pricing $2/$10 per MTok. Comparative latency Fast → speed_class 速め. Output $10 → cost_class 中. Claude Code Anthropic API sonnet alias → Sonnet 5 (v2.1.197+). T3 menus use eri-confirmed spellings (Extra High / Ultrathink), not normalized to official xhigh.
claude-fable-5-1 · claude-fable-5-1
Demanding reasoning and long-horizon agentic work; stronger long-running agentic coding, multistep research, and document/spreadsheet/slide work (Anthropic Fable 5.1 overview)
Adaptive thinking always on; API effort default high; all five effort levels. Pricing $10/$50 per MTok. Comparative latency Slower → speed_class 遅め. Output $50 → cost_class 特高. Released 2026-09-01. Max output 128K. T3 labels/menus unknown (eri 2026-09-17).
chatgpt-5-6-sol · gpt-5.6-sol
GPT-5.6 flagship for complex professional work; coding, knowledge work, computer use (OpenAI models + GPT-5.6 post)
Alias gpt-5.6 → Sol. Seed catalog id chatgpt-5-6-sol kept. reasoning.effort none|low|medium|high|xhigh|max; default medium. Pricing $4/$20 per MTok (promo through at least 2026-11-21). Model page latency badge Fast → speed_class 速め. Output $20 → cost_class 高. Function calling yes. Distinct from GPT-6 Sol (gpt-6-sol). T3 labels/menus unknown (eri 2026-09-17).
chatgpt-5-6-terra · gpt-5.6-terra
Balances intelligence and cost; roughly the mini tier of the GPT-5.6 family (OpenAI GPT-5.6 Terra model page)
Official Fast latency badge → speed_class 速め. Pricing $2/$12 per MTok (short context); output $12 → cost_class 中. Long-context output $18 would be 高 — used short-context base. Function calling yes. reasoning.effort none|low|medium|high|xhigh|max; default medium. Models overview Choosing a model now points balance/cost defaults to GPT-6 Sol/Luna; GPT-5.6 Terra remains active. T3 labels/menus unknown (not in eri 2026-09-17 screen notes).
chatgpt-5-6-luna · gpt-5.6-luna
Cost-sensitive high-volume workloads; roughly the nano tier of the GPT-5.6 family (OpenAI GPT-5.6 Luna model page)
Official Fast latency badge → speed_class 速め. Pricing $0.20/$1.20 per MTok (short context); output $1.20 → cost_class 低. Function calling yes. reasoning.effort none|low|medium|high|xhigh|max; default medium. coding left unknown (no official coding-strength label on model page). Distinct from GPT-6 Luna (gpt-6-luna). Models overview Choosing a model now points cost-sensitive defaults to GPT-6 Luna; GPT-5.6 Luna remains active. T3 labels/menus unknown (not in eri 2026-09-17 screen notes).
chatgpt-6-astra · gpt-6-astra
Most capable OpenAI model for hardest end-to-end reasoning, coding, computer use, research (OpenAI models + GPT-6 Astra page)
Seed catalog id chatgpt-6-astra kept. reasoning.effort low|medium|high|xhigh|max (none unsupported). Official model page does not state API default effort → effort_default unknown (T3 Reasoning default Medium is UI-only). Pricing $10/$50 per MTok; Fast mode 2x price. Model page latency badge Fast → speed_class 速め (verified 2026-09-28). Output $50 → cost_class 特高. Function calling yes. first_seen empty pending authoritative launch-day stamp. T3 menus use eri-confirmed spellings (Extra High / Ultra).
chatgpt-6-sol · gpt-6-sol
Built for complex coding and agentic workflows; GPT-6 mid-tier balancing intelligence and cost (OpenAI GPT-6 Sol model page + models overview)
Distinct from GPT-5.6 Sol (gpt-5.6-sol). reasoning.effort none|low|medium|high|xhigh|max; default medium. Pricing $2/$10 per MTok. Model page latency badge Fast → speed_class 速め. Output $10 → cost_class 中. Function calling yes. Announced 2026-09-22. OpenAI models overview Choosing a model now recommends GPT-6 Sol to balance intelligence and cost (was GPT-5.6 Terra in earlier guidance). T3 labels/menus unknown (not yet confirmed by eri).
chatgpt-6-luna · gpt-6-luna
Most efficient GPT-6 model for focused, high-volume tasks (OpenAI GPT-6 Luna model page + models overview)
Distinct from GPT-5.6 Luna (gpt-5.6-luna). reasoning.effort none|low|medium|high|xhigh|max; default medium. Pricing $0.10/$0.50 per MTok; output $0.50 → cost_class 低. Model page latency badge Fast → speed_class 速め (verified 2026-09-28). OpenAI models overview now recommends GPT-6 Luna for cost-sensitive high-volume workloads (alongside GPT-6 Sol for balance). Function calling yes. coding left unknown (efficiency/high-volume wording, no coding-strength label). Announced 2026-09-22. T3 labels/menus unknown (not yet confirmed by eri).
gemini-3-8-flash · gemini-3.8-flash
Most intelligent Flash: long-horizon software engineering, autonomous agents, complex enterprise workflows (Gemini 3.8 Flash model page)
thinking_level low|medium|high; default medium (minimal unsupported). Intro pricing $0.75/$3.75 per MTok through 2026-12-31 → output $3.75 cost_class 低. No Anthropic-style comparative latency label on model page → speed_class unknown (Flash marketing speed wording not treated as comparative latency enum). Function calling yes. T3: three labels by effort variant; menus none; default High label.
gemini-3-1-pro · gemini-3.1-pro-preview
Advanced reasoning/multimodal Pro; optimized for software engineering and precise tool-using agentic workflows (Gemini 3.1 Pro preview page)
Public endpoint gemini-3.1-pro-preview; status=preview. thinking_level low|medium|high; default high. Pricing $2/$12 per MTok (<=200k); $4/$18 (>200k). cost_class 中 from standard <=200k output $12 (note: >200k output $18 would be 高). speed_class unknown. Function calling + thinking yes. T3: Legacy models has High and Low two entries; exact labels unknown (eri 2026-09-17).
grok-4-6 · grok-4.6
Frontier coding, agentic tasks, and knowledge work; long-running agents (xAI Grok 4.6 docs + news)
reasoning_effort low|medium|high|xhigh; default high. Pricing $2/$6 per MTok (<200k); $4/$12 (>=200k). cost_class 中 from base output $6. No official comparative-latency enum → speed_class unknown (per eri: reset from 速め). Function calling yes. T3 Fast choices quoted On/Off; Fast default unknown. T3 Effort uses Extra High (not xhigh). T3: ウニ様=Grok Build を eri 追記。Grok Build 側 Effort/Fast メニューは未確認のため空。Successor Grok 4.7 now recommended for code/chat (xAI models docs).
grok-4-7 · grok-4.7
Frontier coding, agentic tasks, and knowledge work; longer difficult tasks and stronger self-checks; xAI recommended model for code and chat (xAI Grok 4.7 docs + news)
reasoning_effort low|medium|high|xhigh; default high. Pricing $2/$6 per MTok (<200k); $4/$12 (>=200k) — same as Grok 4.6. cost_class 中 from base output $6. No official comparative-latency enum → speed_class unknown. Function calling yes. Announced 2026-09-21. Fast variant 2x speed / 2x price mentioned in news. T3 labels/menus unknown (not yet confirmed by eri).