openclacky 1.5.12 → 1.5.13

data/lib/clacky/utils/model_pricing.rb

@@ -7,6 +7,23 @@ module Clacky
     # Pricing per 1M tokens (MTok) in USD
     # All pricing is based on official API documentation
     PRICING_TABLE = {
+      # Claude Fable 5.1 — same input/output as Fable 5, but cache read is
+      # $0.25/MTok (vs $1.00 for Fable 5). Source: Anthropic pricing table.
+      "claude-fable-5-1" => {
+        input: {
+          default: 10.00,              # $10/MTok for prompts ≤ 200K tokens
+          over_200k: 10.00
+        },
+        output: {
+          default: 50.00,             # $50/MTok for prompts ≤ 200K tokens
+          over_200k: 50.00
+        },
+        cache: {
+          write: 12.50,               # $12.50/MTok cache write (5-min tier)
+          read: 0.25                  # $0.25/MTok cache read
+        }
+      },
+
       # Claude 4.5 models - tiered pricing based on prompt length
       "claude-fable-5" => {
         input: {
@@ -408,6 +425,26 @@ module Clacky
         }
       },
 
+      # Gemini 3.8 Flash. Flat pricing, 1M context, 64K max output.
+      # Same list price as 3.7/3.6 Flash; intro rate $0.75/$3.75 through
+      # 2026-12-31 before settling at $1.50/$7.50 (list price shown).
+      # Cache write billed at input rate (Vertex doesn't expose a separate
+      # cache-write charge in the OpenAI shim usage response).
+      "gemini-3.8-flash" => {
+        input: {
+          default: 1.50,
+          over_200k: 1.50
+        },
+        output: {
+          default: 7.50,
+          over_200k: 7.50
+        },
+        cache: {
+          write: 1.50,
+          read: 0.15
+        }
+      },
+
       # Gemini 3.7 Flash. Flat pricing, 1M context, 64K max output.
       # Same list price as 3.6 Flash; both carry an intro rate of $0.75/$3.75
       # through 2026-12-31 before settling at $1.50/$7.50 (list price shown).
@@ -617,6 +654,18 @@ module Clacky
       # GLM-5.3 shares GLM-5.2's base model (all gains are post-training per
       # Z.ai's release notes) and Z.ai's pricing page does not list a separate
       # GLM-5.3 row yet, so it is billed at the GLM-5.2 flat rate.
+      # GLM-5.3-Flash: GLM-5's first natively-multimodal model (image/video/
+      # file input). Listed at the non-promotional Z.ai rates — input $0.15,
+      # output $0.50, cache read $0.03 per 1M tokens; the 50%-off launch promo
+      # ($0.075 / $0.25 / $0.015, ends 2026-09-09) is ignored per the
+      # "displayed ≤ actual" rule. Cache write bills at the input miss rate
+      # (storage is "Limited-time Free").
+      "glm-5.3-flash" => {
+        input:  { default: 0.15, over_200k: 0.15 },
+        output: { default: 0.50, over_200k: 0.50 },
+        cache:  { write: 0.15, read: 0.03 }
+      },
+
       "glm-5.3" => {
         input:  { default: 1.40, over_200k: 1.40 },
         output: { default: 4.40, over_200k: 4.40 },
@@ -724,6 +773,15 @@ module Clacky
       #   - cache.read  = official explicit-cache-hit price.
       #   - When a model has NO published explicit-cache price (e.g. qwen3.6-27b,
       #     qwen-plus-latest), cache.write/read fall back to the input rate.
+      # qwen3.8-max: NOT tiered (single flat tier). Reuses qwen3.7-max's
+      #   international list rates (2.5 / 7.5 / 3.125 / 0.25) — the billing
+      #   page hasn't listed qwen3.8-max yet, so this is a safe upper bound.
+      "qwen3.8-max" => {
+        input:  { default: 2.5, over_200k: 2.5 },
+        output: { default: 7.5, over_200k: 7.5 },
+        cache:  { write: 3.125, read: 0.25 }
+      },
+
       # qwen3.7-max: NOT tiered (single flat tier per Alibaba's definition).
       #   List price: input 2.5, output 7.5, explicit write 3.125, explicit read 0.25.
       "qwen3.7-max" => {
@@ -908,6 +966,8 @@ module Clacky
         # Support both dot and dash separators (e.g., "4.5", "4-5", "4-6")
         # Also handles Bedrock cross-region prefixes (e.g. "jp.anthropic.claude-sonnet-4-6")
         case model
+        when /claude.*fable.*5[.-]1/i
+          "claude-fable-5-1"
         when /claude.*fable.*5/i
           "claude-fable-5"
         # Claude Sonnet 5 / Opus 5 (2026) — anchored on the literal "sonnet-5"
@@ -975,6 +1035,8 @@ module Clacky
         # (mainland bigmodel.cn vs intl z.ai) the user configured.
         # Strict anchored match so unrelated strings like "glm-5-x-foo"
         # don't silently borrow a nearby model's rate.
+        when /^glm-5\.3-flash$/i
+          "glm-5.3-flash"
         when /^glm-5\.3$/i
           "glm-5.3"
         when /^glm-5\.2$/i
@@ -998,10 +1060,12 @@ module Clacky
           "minimax-m2.5"
 
         # Qwen (Alibaba DashScope) — strict anchored match per registered
-        # model id in providers.rb. qwen3.7-* is the latest flagship line;
-        # qwen3.6-* are the previous generation; qwen-plus-latest is the
-        # rolling alias for the latest Qwen-Plus release; qwen3-vl-plus is
-        # the multimodal SKU (replaces the retired qwen-vl-plus/max).
+        # model id in providers.rb. qwen3.8-max is the latest flagship;
+        # qwen3.7-* / qwen3.6-* are previous generations; qwen-plus-latest
+        # is the rolling alias for the latest Qwen-Plus release; qwen3-vl-plus
+        # is the multimodal SKU (replaces the retired qwen-vl-plus/max).
+        when /^qwen3\.8-max$/i
+          "qwen3.8-max"
         when /^qwen3\.7-max$/i
           "qwen3.7-max"
         when /^qwen3\.7-plus$/i
@@ -1029,6 +1093,8 @@ module Clacky
           "gemini-3.6-flash"
         when /^or-gemini-3-7-flash$/i, /^gemini-3\.7-flash$/i
           "gemini-3.7-flash"
+        when /^or-gemini-3-8-flash$/i, /^gemini-3\.8-flash$/i
+          "gemini-3.8-flash"
 
         # OpenAI GPT-5.x models - match various dashed/dotted/compact forms
         # (e.g. "gpt-5.5", "gpt-5-5", "gpt5.5", "gpt55")

data/lib/clacky/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Clacky
-  VERSION = "1.5.12"
+  VERSION = "1.5.13"
 end