Daily Neural Digest
Back to LLM Cost Changes
llm costsllm-costsapi-pricingopenrouter

LLM API price changes — what moved in the catalogue

Daily diff of 441 LLM API models tracked from OpenRouter /api/v1/models: price changes, new and removed models, context windows. Read from the API, not written by a model.

FTL LabSeptember 21, 20264 min read705 words

What this page is

We snapshot the full LLM API catalogue behind OpenRouter /api/v1/models once a day and publish what changed: input and output prices, context windows, models that appeared, models that vanished. 441 models are tracked. Every figure below is read from the API response and compared by script — no language model writes or interprets any number on this page, so there is nothing to hallucinate.

A price move is reported only if it is both ≥ 5 % and ≥ $0.02 per million tokens; a context-window move only if it is ≥ 10 %. Appearances and removals are always reported: a model leaving a catalogue breaks a deployment, and a model joining it can cut your bill. 24 of the tracked models are free to call.

Why it matters: list prices for the same capability moved by a factor of more than a hundred across this catalogue, and vendors rarely announce a quiet repricing of an older model. If you route traffic by name, the page tells you when the name you route to stopped being the cheapest or disappeared.

Changes recorded (1 comparison day(s))

2026-09-21

  • Removed from the catalogue — deepseek/deepseek-v4-flash-0731:free
  • Price sortie — deepseek/deepseek-v4-flash-0731: $0.08 → $0.16 per M tokens (+100.0 %)
  • Price sortie — ~deepseek/deepseek-v4-flash-latest: $0.08 → $0.16 per M tokens (+100.0 %)
  • Price entrée — deepseek/deepseek-v4-flash: $0.0372 → $0.0899 per M tokens (+141.7 %)
  • Price sortie — deepseek/deepseek-v4-flash: $0.0745 → $0.1797 per M tokens (+141.2 %)
  • Price entrée — ibm-granite/granite-4.2-8b: $0.06 → $0.1 per M tokens (+66.7 %)
  • Price sortie — ibm-granite/granite-4.2-8b: $0.25 → $0.15 per M tokens (-40.0 %)
  • Price sortie — ~deepseek/deepseek-flash-latest: $0.52 → $0.48 per M tokens (-7.7 %)
  • Price entrée — qwen/qwen3.6-35b-a3b: $0.1 → $0.15 per M tokens (+50.0 %)
  • Price sortie — qwen/qwen3.6-35b-a3b: $0.9 → $1.0 per M tokens (+11.1 %)
  • Price sortie — meta-llama/llama-4-maverick: $0.6525 → $0.8 per M tokens (+22.6 %)
  • Price entrée — qwen/qwen3.8-27b: $0.42 → $0.2 per M tokens (-52.4 %)
  • Price sortie — qwen/qwen3.8-27b: $3.0 → $2.5 per M tokens (-16.7 %)
  • Price entrée — deepseek/deepseek-v4.1-flash: $0.15 → $0.3 per M tokens (+100.0 %)
  • Price sortie — deepseek/deepseek-v4.1-flash: $0.6 → $1.2 per M tokens (+100.0 %)
  • Price entrée — meta/muse-glimmer-30b: $0.35 → $0.3 per M tokens (-14.3 %)
  • Price sortie — meta/muse-glimmer-30b: $1.5 → $1.2 per M tokens (-20.0 %)
  • Price entrée — z-ai/glm-5.2: $0.5544 → $0.6496 per M tokens (+17.2 %)
  • Price sortie — z-ai/glm-5.2: $1.7424 → $2.0416 per M tokens (+17.2 %)
  • Price entrée — ~deepseek/deepseek-pro-latest: $0.5782 → $0.7 per M tokens (+21.1 %)
  • Price sortie — ~deepseek/deepseek-pro-latest: $1.7345 → $2.88 per M tokens (+66.0 %)
  • Price entrée — meta-llama/llama-3.1-70b-instruct: $0.4 → $0.72 per M tokens (+80.0 %)
  • Price sortie — meta-llama/llama-3.1-70b-instruct: $0.4 → $0.72 per M tokens (+80.0 %)
  • Price entrée — ~z-ai/glm-latest: $0.8442 → $0.7735 per M tokens (-8.4 %)
  • Price sortie — ~z-ai/glm-latest: $2.6532 → $2.431 per M tokens (-8.4 %)
  • Price entrée — deepseek/deepseek-v4-pro: $0.4223 → $0.9483 per M tokens (+124.6 %)
  • Price sortie — deepseek/deepseek-v4-pro: $0.8446 → $1.8966 per M tokens (+124.6 %)
  • Price entrée — deepseek/deepseek-v4-pro-0813: $0.5782 → $1.32 per M tokens (+128.3 %)
  • Price sortie — deepseek/deepseek-v4-pro-0813: $1.7345 → $3.96 per M tokens (+128.3 %)

How to read a change

  • Removed is the one that hurts: your calls start failing, and no vendor email arrives for a model that was never announced.
  • Price moves are per million tokens and exclude caching discounts, batch discounts and provider routing differences — treat them as the list price signal, not your invoice.
  • Context window moves usually mean a new version was published under the same name; re-read behaviour as well as price before switching.

Sources and method

  • Catalogue read from OpenRouter /api/v1/models once per day, stored verbatim in data/history/pricing/.
  • Comparison performed by pipelines/pricing_changes.py; thresholds are fixed in code and shown above.
  • One snapshot per day; if a collector fails, the guard refuses a truncated snapshot so a temporary outage never appears as hundreds of models being removed.
  • Source of record: https://openrouter.ai/api/v1/models
llm-costsapi-pricingopenrouterftl-lab
Share this article:

Was this article helpful?

Let us know to improve our AI generation.