Modelle
Was sich zwischen Versionen geändert hat
Ausgewählte Anthropic-Modelle. Prüfdaten stehen pro Eintrag; ältere Datensätze behalten ihr ursprüngliches Datum. Die Tabelle zeigt Standard-API-Limits.
Aktuell
| Modell | ID | Kontext | Max. Ausgabe | Eingabe | Ausgabe | Denkmodus | Geprüft |
|---|---|---|---|---|---|---|---|
| Claude Fable 5.1 | claude-fable-5-1 | 1M | 128K | $10 | $50 | always-on | Preis: 2026-09-08 |
| Claude Opus 5 | claude-opus-5 | 1M | 128K | $5 | $25 | adaptive-default | |
| Claude Sonnet 5 | claude-sonnet-5 | 1M | 128K | $2 | $10 | adaptive-default | Preis: 2026-09-08 Verhalten: 2026-09-08 |
| Claude Haiku 4.5 | claude-haiku-4-5 | 200K | 64K | $1 | $5 | budget-tokens |
Älter
| Modell | ID | Kontext | Max. Ausgabe | Eingabe | Ausgabe | Denkmodus | Geprüft |
|---|---|---|---|---|---|---|---|
| Claude Fable 5 | claude-fable-5 | 1M | 128K | $10 | $50 | always-on | |
| Claude Opus 4.8 | claude-opus-4-8 | 1M | 128K | $5 | $25 | adaptive-opt-in | |
| Claude Opus 4.7 | claude-opus-4-7 | 1M | 128K | $5 | $25 | adaptive-opt-in | |
| Claude Opus 4.6 | claude-opus-4-6 | 1M | 128K | $5 | $25 | adaptive-recommended | |
| Claude Sonnet 4.6 | claude-sonnet-4-6 | 1M | 128K | $3 | $15 | adaptive-recommended |
Wann sinnvoll
- Claude Fable 5.1
- Evaluate when your established baseline misses demanding reasoning or long-running tasks.
- Adaptive thinking is always on; the documented default effort is high.
- Context and output figures are standard API limits; other processing options have separate conditions.
- Claude Fable 5
- The hardest reasoning and the longest agentic runs, when you have already tried Opus 5 and the ceiling was the problem.
- Thinking cannot be disabled; depth is controlled with effort.
- Requires 30-day data retention — not available under zero data retention.
- Single requests on hard tasks can run for many minutes.
- Claude Opus 5
- The default. Start here and only move if you have a reason.
- Thinking is on by default — unlike Opus 4.8 and 4.7.
- Prefer low or medium effort over disabling thinking.
- Claude Opus 4.8
- Existing workloads already tuned for it.
- Omitting the thinking parameter runs without thinking.
- Claude Opus 4.7
- Nothing new. Migrate to Opus 5.
- Claude Opus 4.6
- Nothing new. Migrate to Opus 5.
- The last model where budget_tokens still works, as a transitional escape hatch only.
- Sampling parameters are still accepted here.
- Claude Sonnet 5
- High volume where Opus-tier quality is not the constraint.
The launch rate became the standard rate. The previously announced September increase was cancelled; see the current pricing source.
- Does not support mid-conversation system messages.
- Claude Sonnet 4.6
- Nothing new. Migrate to Sonnet 5.
- Claude Haiku 4.5
- Cheap, fast, high-volume work — classification, extraction, and sub-agents doing mechanical reading.
- The only listed current Anthropic model with a 200K context window.
- Effort is not supported; thinking still uses budget_tokens.