Prompt-caching cost optimization
Cut spend on long system prompts, tool schemas, and RAG prefixes via provider prompt caching / context caching with predictable hit rates.
At a glance — Oct 2026
Updated 2026-10-27 · forward this page: /jtbds/prompt-caching-optimization#briefScores as of 2026-10-19 (8 days old).
- Depot Pick
- Anthropic · 8.00/10 for this job (curated estimate, high confidence)
pricePerf + DX of cache controls + hit-rate observability. Anthropic prompt caching is the clearest agent-facing story; OpenAI/Gemini close. Confidence: high.
- Runner-up
- OpenAI · 7.72/10
Scores higher than the pick on: Latency / reliability, Developer & agent DX, Maturity / track record.
- When not to use the pick
- Workloads are short-prefix high-variance (cache thrash), or gateway-level cache across vendors (Portkey/CF AI Gateway) matters more than first-party cache.
- Lock-in / migration
- The pick has lower stored lock-in risk than runner-up OpenAI (6.5 vs 7.2).
- Sourced price (usd_per_million_input_tokens)
Pick's lowest sourced SKU: Haiku 5.5 (prompts up to 100K tokens) $0.1 (usd_per_million_input_tokens) — source, as of 2026-10-27, re-checked 2026-10-27
Cheapest matching SKU: Amazon Bedrock Gemma 3 4B $0.04 — source, as of 2026-10-03, re-checked 2026-10-27. Not the pick; not quality-adjusted.
Price history: first seen 2026-10-03 · last changed no change recorded · history JSON
- Free option
- No free resource lists this job. A cheap paid SKU is not a free option.
- Failure modes to watch
- Runaway spend: Set a budget and a tool-call cap per task. Alert when spend jumps. Stop the run at the cap instead of summarizing a partial trace as success.
Same-job, same-unit comparisons: Anthropic vs OpenAI · Anthropic vs Google AI (Gemini) · Anthropic vs Amazon Bedrock
Cite: Agent Provider Depot, Depot Pick for prompt-caching-optimization: anthropic, scores as of 2026-10-19. Agents: GET /api/decide?jtbd=prompt-caching-optimization
All sourced prices, cheapest first
Anthropic
pricePerf + DX of cache controls + hit-rate observability. Anthropic prompt caching is the clearest agent-facing story; OpenAI/Gemini close. Confidence: high.
Pricing (directional): Sonnet 5 ~$2/$10 per 1M; Opus ~$5/$25; Haiku 4.5 ~$1/$5; Fable higher. Prompt caching + batch. Managed Agents session-hour fees. US-only inference ~1.1x.
Lock-in risk: higher = more lock-in (worse). Overall already inverts this dimension.
Depot default for serious coding and production agents. Prefer Sonnet for cost/quality; reserve Opus for hard reasoning. Pair with Bedrock/Vertex if procurement demands cloud wrappers.
Quote for Prompt-caching cost optimization
Depot Pick stays Anthropic. Prices below are sourced SKU rows only (as of 2026-10-27). Input and output are not blended. GET /api/quote?jtbd=prompt-caching-optimization
Pick is not a price winner. anthropic sourced input $/1M is 0.1; cheapest among candidates is amazon-bedrock at 0.04.
- Fable 5.1 usd_per_million_input_tokens $10 source as of 2026-10-27
- Haiku 4.5 usd_per_million_input_tokens $1 source as of 2026-10-27
- Haiku 5.5 (prompts over 100K tokens) usd_per_million_input_tokens $0.5 source as of 2026-10-27
- Haiku 5.5 (prompts up to 100K tokens) usd_per_million_input_tokens $0.1 source as of 2026-10-27
- Opus 5.5 usd_per_million_input_tokens $4 source as of 2026-10-27
- Sonnet 5.5 usd_per_million_input_tokens $2 source as of 2026-10-27
- Fable 5.1 usd_per_million_output_tokens $50 source as of 2026-10-27
- Haiku 4.5 usd_per_million_output_tokens $5 source as of 2026-10-27
- Haiku 5.5 (prompts over 100K tokens) usd_per_million_output_tokens $2.5 source as of 2026-10-27
- Haiku 5.5 (prompts up to 100K tokens) usd_per_million_output_tokens $0.5 source as of 2026-10-27
- Opus 5.5 usd_per_million_output_tokens $20 source as of 2026-10-27
- Sonnet 5.5 usd_per_million_output_tokens $10 source as of 2026-10-27
- Code execution additional hours usd_per_hour $0.05 source as of 2026-10-27
- Managed Agents session runtime usd_per_session_hour $0.08 source as of 2026-10-27
- Web search usd_per_1k_searches $10 source as of 2026-10-27
0 of 5 candidates have no sourced price row. 1 are excluded from the Input $/1M comparison (chat_completion_input). min sourced input SKU among candidates; not quality-adjusted
Sourced Input $/1M for Prompt-caching cost optimization
Candidates for this job that already have a matching sourced price (chat_completion_input), cheapest first. As of 2026-10-27. Each figure is one SKU in that unit, not a blend of input and output, not a cross-modal rate, and not a Depot score. min sourced input SKU among candidates; not quality-adjusted The Depot Pick is not a price winner.
| Provider | Input $/1M | SKU | Source | As of |
|---|---|---|---|---|
| Amazon Bedrock | $0.04 | Gemma 3 4B | pricing page | 2026-10-03 |
| AnthropicDepot Pick | $0.1 | Haiku 5.5 (prompts up to 100K tokens) | pricing page | 2026-10-27 |
| OpenAI | $0.1 | gpt-6-luna | pricing page | 2026-10-03 |
| Google AI (Gemini) | $0.3 | gemini-3.5-flash-lite | pricing page | 2026-10-21 |
Runners-up
OpenAI
Default when you want maximum capability and ecosystem glue. Multi-home critical paths behind OpenAI-compat…
Google AI (Gemini)
Best value frontier stack and multimodal king. Use Flash for high-volume agent loops; Pro when quality…
Portkey
Depot pick for production AI gateway (fallbacks, virtual keys, budgets). OpenRouter when you want a model…
Amazon Bedrock
Enterprise multi-model switchboard. Great when AWS is mandatory; mediocre when you just want the best Claude…
When NOT to use the top pick
Workloads are short-prefix high-variance (cache thrash), or gateway-level cache across vendors (Portkey/CF AI Gateway) matters more than first-party cache.
Full ranking (score table)
Ranked with this JTBD's weights. Scores are curated estimates unless marked otherwise. Agents: prefer /api/decide then this table for tie-breaks.
| # | Provider | Overall | Fit | $/perf | Latency | DX | As of |
|---|---|---|---|---|---|---|---|
| 1 | Google AI (Gemini) | 8.4 | 9.2 | 8.8 | 8.5 | 8.2 | 2026-10-19 |
| 2 | Portkey | 8.2 | 8.8 | 8.2 | 8.6 | 9.0 | 2026-09-14 |
| 3 | AnthropicDepot Pick | 8.0 | 9.7 | 6.8 | 8.7 | 9.2 | 2026-10-19 |
| 4 | OpenAI | 7.7 | 9.6 | 6.2 | 8.8 | 9.5 | 2026-09-14 |
| 5 | Amazon Bedrock | 7.3 | 8.8 | 6.0 | 9.2 | 6.8 | 2026-09-14 |
Score methodology for this job
pricePerf + DX of cache controls + hit-rate observability. Anthropic prompt caching is the clearest agent-facing story; OpenAI/Gemini close. Confidence: high.
capabilityFit: 18%pricePerformance: 32%latencyReliability: 14%developerDx: 12%privacyPosture: 8%lockInRisk: 8%maturity: 8%