Undominated.ai

Context cliffs

Every published context-tier step function in the catalogue. Ladders keep every rung — a flattened ladder once mispriced five models by up to 2×.

5 of 71 carry more than one rung. Multipliers are input $/M at the rung ÷ headline input $/M. Condition: OpenRouter-published tiers in the catalogue payload.

ModelHeadline inRungsMax ×
Qwen3.7 FlashCheck
  1. > 32K → $0.100/M in (3.33×) first rung, not the top
  2. > 256K → $0.200/M in (6.67×)
$0.03026.67×
Qwen3.6 FlashCheck
  1. > 256K → $0.750/M in (4×)
$0.18814×
Qwen3.6 PlusCheck
  1. > 256K → $1.30/M in (4×)
$0.32514×
Qwen Plus 0728Check
  1. > 256K → $0.780/M in (3×)
$0.26013×
Qwen-PlusCheck
  1. > 256K → $0.780/M in (3×)
$0.26013×
Qwen3 Coder PlusCheck
  1. > 32K → $1.17/M in (1.8×) first rung, not the top
  2. > 128K → $1.95/M in (3×)
$0.65023×
Qwen3.7 PlusCheck
  1. > 256K → $0.960/M in (3×)
$0.32013×
Qwen3 Coder FlashCheck
  1. > 32K → $0.325/M in (1.67×) first rung, not the top
  2. > 128K → $0.520/M in (2.67×)
$0.19522.67×
Qwen3 MaxCheck
  1. > 32K → $1.56/M in (2×) first rung, not the top
  2. > 128K → $1.95/M in (2.5×)
$0.78022.5×
Qwen3 Max ThinkingCheck
  1. > 32K → $1.56/M in (2×) first rung, not the top
  2. > 128K → $1.95/M in (2.5×)
$0.78022.5×
Claude Sonnet 4Check
  1. > 200K → $6.00/M in (2×)
$3.0012×
Claude Sonnet 4.5Check
  1. > 200K → $6.00/M in (2×)
$3.0012×
Claude Sonnet 4.5 (batch)
  1. > 200K → $3.00/M in (2×)
$1.5012×
Fugu UltraCheck
  1. > 272K → $10.00/M in (2×)
$5.0012×
Fugu Ultra v2Check
  1. > 272K → $10.00/M in (2×)
$5.0012×
Gemini 2.5 ProCheck
  1. > 200K → $2.50/M in (2×)
$1.2512×
Gemini 2.5 Pro (batch)
  1. > 200K → $1.25/M in (2×)
$0.62512×
Gemini 2.5 Pro Preview 06-05Check
  1. > 200K → $2.50/M in (2×)
$1.2512×
Gemini 3.1 Pro PreviewCheck
  1. > 200K → $4.00/M in (2×)
$2.0012×
Gemini 3.1 Pro Preview (batch)
  1. > 200K → $2.00/M in (2×)
$1.0012×
Gemini 3.1 Pro Preview Custom ToolsCheck
  1. > 200K → $4.00/M in (2×)
$2.0012×
GPT-5.4Check
  1. > 272K → $5.00/M in (2×)
$2.5012×
GPT-5.4 (batch)
  1. > 272K → $2.50/M in (2×)
$1.2512×
GPT-5.4 ProCheck
  1. > 272K → $60.00/M in (2×)
$30.0012×
GPT-5.4 Pro (batch)
  1. > 272K → $30.00/M in (2×)
$15.0012×
GPT-5.5Check
  1. > 272K → $10.00/M in (2×)
$5.0012×
GPT-5.5 (batch)
  1. > 272K → $5.00/M in (2×)
$2.5012×
GPT-5.5 ProCheck
  1. > 272K → $60.00/M in (2×)
$30.0012×
GPT-5.5 Pro (batch)
  1. > 272K → $30.00/M in (2×)
$15.0012×
GPT-5.6 LunaCheck
  1. > 272K → $0.400/M in (2×)
$0.20012×
GPT-5.6 Luna (batch)
  1. > 272K → $0.200/M in (2×)
$0.10012×
GPT-5.6 Luna ProCheck
  1. > 272K → $0.400/M in (2×)
$0.20012×
GPT-5.6 Luna Pro (batch)
  1. > 272K → $0.200/M in (2×)
$0.10012×
GPT-5.6 SolCheck
  1. > 272K → $8.00/M in (2×)
$4.0012×
GPT-5.6 Sol (batch)
  1. > 272K → $2.00/M in (2×)
$1.0012×
GPT-5.6 Sol ProCheck
  1. > 272K → $8.00/M in (2×)
$4.0012×
GPT-5.6 Sol Pro (batch)
  1. > 272K → $2.00/M in (2×)
$1.0012×
GPT-5.6 TerraCheck
  1. > 272K → $4.00/M in (2×)
$2.0012×
GPT-5.6 Terra (batch)
  1. > 272K → $2.00/M in (2×)
$1.0012×
GPT-5.6 Terra ProCheck
  1. > 272K → $4.00/M in (2×)
$2.0012×
GPT-5.6 Terra Pro (batch)
  1. > 272K → $2.00/M in (2×)
$1.0012×
GPT-6 AstraCheck
  1. > 272K → $20.00/M in (2×)
$10.0012×
GPT-6 Astra (batch)
  1. > 272K → $10.00/M in (2×)
$5.0012×
GPT-6 Astra ProCheck
  1. > 272K → $20.00/M in (2×)
$10.0012×
GPT-6 Astra Pro (batch)
  1. > 272K → $10.00/M in (2×)
$5.0012×
GPT-6 LunaCheck
  1. > 272K → $0.200/M in (2×)
$0.10012×
GPT-6 Luna (batch)
  1. > 272K → $0.100/M in (2×)
$0.05012×
GPT-6 Luna ProCheck
  1. > 272K → $0.200/M in (2×)
$0.10012×
GPT-6 Luna Pro (batch)
  1. > 272K → $0.100/M in (2×)
$0.05012×
GPT-6 SolCheck
  1. > 272K → $4.00/M in (2×)
$2.0012×
GPT-6 Sol (batch)
  1. > 272K → $2.00/M in (2×)
$1.0012×
GPT-6 Sol ProCheck
  1. > 272K → $4.00/M in (2×)
$2.0012×
GPT-6 Sol Pro (batch)
  1. > 272K → $2.00/M in (2×)
$1.0012×
GPT-6.1 SolCheck
  1. > 272K → $4.00/M in (2×)
$2.0012×
GPT-6.1 Sol ProCheck
  1. > 272K → $4.00/M in (2×)
$2.0012×
Grok 4.20Check
  1. > 200K → $2.50/M in (2×)
$1.2512×
Grok 4.20 Multi-AgentCheck
  1. > 200K → $2.50/M in (2×)
$1.2512×
Grok 4.3Check
  1. > 200K → $2.50/M in (2×)
$1.2512×
Grok 4.3 (batch)
  1. > 200K → $2.00/M in (2×)
$1.0012×
Grok 4.5Check
  1. > 200K → $4.00/M in (2×)
$2.0012×
Grok 4.6Check
  1. > 200K → $4.00/M in (2×)
$2.0012×
Grok 4.7Check
  1. > 200K → $4.00/M in (2×)
$2.0012×
Grok Build 0.1Check
  1. > 200K → $2.00/M in (2×)
$1.0012×
Seed 1.6Check
  1. > 128K → $0.500/M in (2×)
$0.25012×
Seed-2.0-CodeCheck
  1. > 128K → $1.00/M in (2×)
$0.50012×
Seed-2.0-LiteCheck
  1. > 128K → $0.500/M in (2×)
$0.25012×
Seed-2.0-MiniCheck
  1. > 128K → $0.200/M in (2×)
$0.10012×
Qwen3.6 Max PreviewCheck
  1. > 128K → $1.58/M in (1.54×)
$1.0311.54×
Seed 1.6 FlashCheck
  1. > 128K → $0.100/M in (1.33×)
$0.07511.33×
Qwen3.5 Plus 2026-02-15Check
  1. > 256K → $0.325/M in (1.25×)
$0.26011.25×
Qwen3.5 Plus 2026-04-20Check
  1. > 256K → $0.375/M in (1.25×)
$0.30011.25×

Questions this page answers

What is a context cliff?

A published step where input $/M rises after a prompt-length threshold. The headline rate applies through the first threshold.

Why does every rung stay visible?

Flattening a tier ladder to one multiplier once mispriced five models by up to 2×. Each rung is kept so long prompts are priced correctly.

Are output tiers included?

This table lists input-tier ladders from the OpenRouter-published payload. Output-tier cliffs appear when vendors publish them.

Input hashes & method version

Read citation text and BibTeX
71 model entries have published context tiers; 5 preserve multiple rungs. Tier rates apply strictly above each threshold; missing cache prices remain unknown. As of 2026-10-06. Lens: Published provider prices in the accepted OpenRouter-derived catalogue. https://undominated.ai/cliffs/
Method: https://undominated.ai/methodology/
Publication: sha256:ecef83f767cd4d2c21766f6a5b4dffa36c06351f23657542d473bc1a030c0932; calculation SHA-256: 5ae4d0bee9b70298dd437b05df87c32acb407040b50bf88240677f295f25955d; permitted inputs: https://undominated.ai/data/publications/manifests/ecef83f767cd4d2c21766f6a5b4dffa36c06351f23657542d473bc1a030c0932.json

@misc{undominated-cliffs-2026-10-06,
  author = {{Undominated.ai}},
  title  = {cliffs evidence — Undominated.ai},
  year   = {2026},
  url    = {https://undominated.ai/cliffs/},
  note   = {as of 2026-10-06; Published provider prices in the accepted OpenRouter-derived catalogue},
  annote = {Publication: sha256:ecef83f767cd4d2c21766f6a5b4dffa36c06351f23657542d473bc1a030c0932; calculation SHA-256: 5ae4d0bee9b70298dd437b05df87c32acb407040b50bf88240677f295f25955d; permitted inputs: https://undominated.ai/data/publications/manifests/ecef83f767cd4d2c21766f6a5b4dffa36c06351f23657542d473bc1a030c0932.json}
}
Evidence & Ask
Markdown for LLMs