savvy_openrouter 0.4.1 → 0.4.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 0b72086a1800026ac12dae2c055b06051b61dec6ea35704d9ebe3a96aac9dee5
-  data.tar.gz: f2d1a708927e1b8fbaf57c10c648a4928cf19714bbae20ee0b4865b918c852b2
+  metadata.gz: 8137e975dade1520d62e20932114c3001f00f068ef088a58e1ffb461c1ba7473
+  data.tar.gz: 6aedce626d20af2e52b5b109c595a7b36a87c1861275d6bcce5d64cf52dfbd7c
 SHA512:
-  metadata.gz: e6cae64c814dd9a1a8162638f28740c05f617085081c32982fce542a1eff413a2010d00977d7a4433886ac2a3d022bc8edeab48693132e53be5ae95b071b03fe
-  data.tar.gz: b50221b09dfce3abbf179bd76d6d8ac951e1382d3bf899cd43008f7cd1f6f2e0d8aa9cc6c08b8deba11e721bb929db33ddabc4cb76c151a1c88c6f87ca9629a0
+  metadata.gz: ca74a1076e38ef38c008d8562744cc88625ca409c93e3ce3474c8413c04cb10fc830a7fc69811578e6d2dee4f1b659b2339319cd151abf2062f446d4df02b0c0
+  data.tar.gz: 30a6e7c41a40710944ad41a2ea9d3886961d4f6ef9e2f6ef7f902c339cb1b7d4e63705a63e2f74163ee06f173cb7ba66f5c2e48c09d20829838f24894e752d9d

data/CHANGELOG.md

@@ -1,5 +1,11 @@
 ## [Unreleased]
 
+## [0.4.2] - 2026-07-14
+
+### Changed
+
+- **`Models#first_ranked_free_text_model`:** passes **`max_price: 0`** to **`GET /models`** (OpenRouter’s free-price filter) while still requiring zero prompt and completion pricing. Temporary free models with an **`expiration_date`** remain eligible when they rank first. Integration expectations updated for current top free ids; smoke chat model switched off the removed `inclusionai/ring-2.6-1t:free` id.
+
 ## [0.4.1] - 2026-05-15
 
 ### Added

data/README.md

@@ -214,14 +214,14 @@ OpenRouter documents the legacy `plugins: [{ id: "web" }]` approach as deprecate
 client.models.list(category: "programming", output_modalities: "text")
 ```
 
-OpenRouter returns models in **curated rank order** within each filtered result set—that order is not “highest `context_length` wins.” To approximate the **top free text model** for a category, call **`first_ranked_free_text_model`**, which keeps API order and returns the **first** model whose **`pricing.prompt`** and **`pricing.completion`** both parse to zero:
+OpenRouter returns models in **curated rank order** within each filtered result set—that order is not “highest `context_length` wins.” To approximate the **top free text model** for a category, call **`first_ranked_free_text_model`**, which calls **`GET /models`** with **`max_price=0`** (and text modality), keeps API order, and returns the **first** model whose **`pricing.prompt`** and **`pricing.completion`** both parse to zero. Temporary free models (those with an **`expiration_date` / “going away” badge) are **included** on purpose when they rank first:
 
 ```ruby
 client.models.first_ranked_free_text_model(category: "programming")
 client.models.first_ranked_free_text_model(category: "roleplay")
 ```
 
-**Use a stored model id for normal traffic.** Resolving the free model calls **`GET /models`** (one request). Each chat turn is **`POST /chat/completions`** (another). If you call `first_ranked_free_text_model` (or `list` and pick) on **every** user message, you pay **two** API calls per turn—list plus chat. Instead, resolve **once** (at deploy, in a Rake task, or on a long TTL), **remember** the returned **`id`** (environment variable, database, cache, or `default_model` in `savvy_openrouter.yml`), and pass that string to **`client.chat.completions(model: ...)`** for ongoing requests. Refresh the stored id when you want to pick up a new “top free” model after OpenRouter changes their list.
+**Use a stored model id for normal traffic.** Resolving the free model calls **`GET /models`** (one request). Each chat turn is **`POST /chat/completions`** (another). If you call `first_ranked_free_text_model` (or `list` and pick) on **every** user message, you pay **two** API calls per turn—list plus chat. Instead, resolve **once** (at deploy, in a Rake task, or on a long TTL), **remember** the returned **`id`** (environment variable, database, cache, or `default_model` in `savvy_openrouter.yml`), and pass that string to **`client.chat.completions(model: ...)`** for ongoing requests. Refresh the stored id when you want to pick up a new “top free” model after OpenRouter changes their list—especially if the chosen model lists an expiration date.
 
 That heuristic stays aligned with OpenRouter’s listing **as long as their ranking and pricing rows stay as they are**; it is not a separate benchmark score from the JSON (there is no `rating` field on each model). For chat-specific knobs such as **`tools`**, **`tool_choice`**, and **`response_format`** (including JSON schema), pass them on **`client.chat.completions`**; global YAML **`defaults`** also merge into embeddings and other resources, so prefer per-call args for tool and schema defaults unless you only use chat endpoints.
 
@@ -282,7 +282,7 @@ bin/setup
 bundle exec rake   # RSpec + RuboCop
 ```
 
-Integration tests live in `spec/integration/` and are tagged `:integration`. When `OPENROUTER_API_KEY` is set, they call the live API (WebMock allows net connect only for those examples). Smoke chat examples use the free model `inclusionai/ring-2.6-1t:free`; **`spec/integration/free_model_rank_spec.rb`** asserts curated “first free” model ids for `programming` and `roleplay` against the live models list (these examples can fail if OpenRouter changes ordering or pricing).
+Integration tests live in `spec/integration/` and are tagged `:integration`. When `OPENROUTER_API_KEY` is set, they call the live API (WebMock allows net connect only for those examples). Smoke chat examples use the free model `nvidia/nemotron-3-super-120b-a12b:free`; **`spec/integration/free_model_rank_spec.rb`** asserts curated “first free” model ids for `programming` and `roleplay` against the live models list (these examples can fail if OpenRouter changes ordering or pricing).
 
 ## Contributing
 

data/lib/savvy_openrouter/resources/models.rb

@@ -13,11 +13,13 @@ module SavvyOpenrouter
         end
       end
 
-      # Uses GET /models with filters, then returns the first model whose prompt + completion pricing are zero.
-      # OpenRouter returns models in curated rank order within a category; first matching free model aligns with
-      # site “top free” picks when combined with output_modalities=text.
+      # Uses GET /models with category + free-price filters, then returns the first model whose prompt and
+      # completion pricing are both zero. OpenRouter returns models in curated rank order; the first free
+      # match aligns with site “top free” picks when combined with output_modalities=text.
+      # Models scheduled for removal (expiration_date) are still eligible — callers often want the current
+      # top free pick even when it is temporary.
       def first_ranked_free_text_model(category:, output_modalities: "text")
-        res = list(category: category, output_modalities: output_modalities)
+        res = list(category: category, output_modalities: output_modalities, max_price: 0)
         data = res[:data] || []
         data.find { |m| free_pricing?(m) }
       end

data/lib/savvy_openrouter/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module SavvyOpenrouter
-  VERSION = "0.4.1"
+  VERSION = "0.4.2"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: savvy_openrouter
 version: !ruby/object:Gem::Version
-  version: 0.4.1
+  version: 0.4.2
 platform: ruby
 authors:
 - Bryan Feller