@@ -44,7 +44,7 @@ Reuse previously-processed prompt prefixes to avoid re-computing the same tokens
4444
4545### Provider Docs
4646
47- - [ Anthropic Prompt Caching] ( https://docs.anthropic.com/en/docs/build-with-claude/prompt-caching ) - 90% discount, 5min/1hr TTL. Minimum cacheable prefix: 4,096 tokens on Opus 4.6/Haiku 4.5, 1,024 on Sonnet 4.6/Opus 4.8.
47+ - [ Anthropic Prompt Caching] ( https://docs.anthropic.com/en/docs/build-with-claude/prompt-caching ) - 90% discount, 5min/1hr TTL. Minimum cacheable prefix: 4,096 tokens on Opus 4.6/Haiku 4.5, 1,024 on Sonnet 4.6/Opus 4.8/Sonnet 5 .
4848- [ Anthropic Caching Announcement] ( https://www.anthropic.com/news/prompt-caching ) - Blog post explaining economics.
4949- [ Anthropic Token-Saving Updates] ( https://www.anthropic.com/news/token-saving-updates ) - Cache-aware rate limits, simplified caching.
5050- [ Anthropic Extended Thinking + Caching] ( https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking ) - Thinking blocks get cached in tool-use loops.
@@ -262,19 +262,20 @@ The [accessibility tree](https://developer.mozilla.org/en-US/docs/Glossary/Acces
262262- [ DeepSeek Pricing] ( https://api-docs.deepseek.com/quick_start/pricing ) - Official DeepSeek pricing.
263263- [ Mistral Pricing] ( https://mistral.ai/pricing ) - Official Mistral pricing.
264264
265- ### Notable Recent Pricing (June 2026)
266-
267- | Model | Input /MTok | Output /MTok | Notes |
268- | --------------------- | ----------- | ------------ | ------------------------------------------------------------------------------------------------------------------------------------------ |
269- | Claude Fable 5 | $10.00 | $50.00 | Anthropic's most capable model; 1M context (June 2026). ** Access suspended June 12 via US export-control directive; restoration pending.** |
270- | Claude Opus 4.8 | $5.00 | $25.00 | 1M context at standard pricing |
271- | GPT-5.5 | $5.00 | $30.00 | OpenAI flagship; 1M context; 90% cached-input discount |
272- | GPT-5.4 | $2.50 | $15.00 | Half the cost of GPT-5.5; 50% Batch API discount |
273- | DeepSeek V4 Flash | $0.14 | $0.28 | Cheapest frontier; 98% cache savings |
274- | DeepSeek V4 Pro | $0.435 | $0.87 | 1M context; thinking + non-thinking modes |
275- | Gemini 3.1 Pro | $2.00 | $12.00 | Preview since Feb 2026; ≤200K context; doubles to $4/$18 above 200K tokens |
276- | Gemini 3.5 Flash | $1.50 | $9.00 | Launched May 19, 2026; 1M context window |
277- | Gemini 2.5 Flash-Lite | $0.10 | $0.40 | Budget option |
265+ ### Notable Recent Pricing (June–July 2026)
266+
267+ | Model | Input /MTok | Output /MTok | Notes |
268+ | --------------------- | ----------- | ------------ | -------------------------------------------------------------------------------------------------------------------------------------------- |
269+ | Claude Fable 5 | $10.00 | $50.00 | Anthropic's most capable model; 1M context (June 2026). Access suspended June 12 via US export-control directive; ** restored July 1, 2026** . |
270+ | Claude Opus 4.8 | $5.00 | $25.00 | 1M context at standard pricing |
271+ | Claude Sonnet 5 | $2.00 | $10.00 | Introductory pricing through Aug 31, 2026 (standard: $3/$15 per MTok); 1M context; most agentic Sonnet; launched June 30, 2026. |
272+ | GPT-5.5 | $5.00 | $30.00 | OpenAI flagship; 1M context; 90% cached-input discount |
273+ | GPT-5.4 | $2.50 | $15.00 | Half the cost of GPT-5.5; 50% Batch API discount |
274+ | DeepSeek V4 Flash | $0.14 | $0.28 | Cheapest frontier; 98% cache savings |
275+ | DeepSeek V4 Pro | $0.435 | $0.87 | 1M context; thinking + non-thinking modes |
276+ | Gemini 3.1 Pro | $2.00 | $12.00 | Preview since Feb 2026; ≤200K context; doubles to $4/$18 above 200K tokens |
277+ | Gemini 3.5 Flash | $1.50 | $9.00 | Launched May 19, 2026; 1M context window |
278+ | Gemini 2.5 Flash-Lite | $0.10 | $0.40 | Budget option |
278279
279280## Prompt Engineering for Efficiency
280281
@@ -407,6 +408,7 @@ The [accessibility tree](https://developer.mozilla.org/en-US/docs/Glossary/Acces
407408| [ Continuous Semantic Caching] ( https://arxiv.org/abs/2604.20021 ) | 2026 | Theory for semantic caching in continuous embedding space; dynamic ε-net + kernel ridge regression |
408409| [ Learning to Draft (LTD)] ( https://arxiv.org/abs/2603.01639 ) | 2026 | RL co-adapts draft+verify policies to optimize true throughput, not acceptance length (ICLR 2026) |
409410| [ DDTree (Block Diffusion)] ( https://arxiv.org/abs/2604.12989 ) | 2026 | Block-diffusion draft tree for speculative decoding; outperforms EAGLE-3 at matched node budget |
411+ | [ Graft] ( https://arxiv.org/abs/2605.20104 ) | 2026 | Training-free prune-then-retrieve framework for speculative decoding draft trees; 5.41× speedup, 21.8% over EAGLE-3 on Qwen3-235B |
410412
411413### Prompt Optimization
412414
0 commit comments