OpenAgents

Commit

inference: declare Gemini 3.7 Flash published rates (refs OpenAgentsInc/openagents#262)

258a25250208 · AtlantisPleb · · parent 52a9102bcbd6 

inference: declare Gemini 3.7 Flash published rates (refs OpenAgentsInc/openagents#262)

Replace the placeholder Gemini rates with Google's Standard paid-tier table.
List rates apply from 2027-01-01; the introductory promotion is half those
rates through that exclusive UTC cutoff. Only declared rates bill (METER-001).

Deploy story

What this commit did to the running system — joined from the forge receipt chain, the part a commit page elsewhere cannot show.

pushed
by user · WAL seq 474 · 2026-08-28T05:14:13.798481Z

Changed files

  • modified config/config.exs
  • modified lib/openagents_web/live/model_catalog_live.ex
  • modified test/openagents/inference/models_test.exs
  • modified test/openagents/inference/pricing_test.exs
  • modified test/openagents_web/controllers/inference_proxy_fallback_test.exs
  • modified test/openagents_web/controllers/model_catalog_controller_test.exs
  • modified test/openagents_web/live/model_catalog_live_test.exs

Diff

7 files changed, +85 -51

config/config.exs modified +21 -21

@@ -139,19 +139,10 @@ config :openagents,

139 139 
  # (including omitting it) for a working figure. `OpenAgents.Inference.Pricing`
140 140 
  # reads that word, and only `:declared` is billable (METER-001).
141 141 
  #
142 
  # Every rate below is a placeholder: no operator has declared any of them, so
143 
  # all of them say `:placeholder` and nothing may bill from a cost they
144 
  # produced. Replacing them is an owner action: enter the provider's rates and
145 
  # set `source: :declared` in the same edit.
146 
  #
147 
  # The GLM 5.3 Flash figures are the ones to be careful with. Its posted rates
148 
  # on 2026-08-26 were $0.075 in, $0.25 out, and $0.015 cached in per million
149 
  # tokens — a limited-time half-price offer that ends 2026-09-09 16:00 UTC,
150 
  # after which list price is $0.15, $0.50, and $0.03. The entry carries list
151 
  # price, because a figure that silently doubles in two weeks is worse than
152 
  # one that is honestly high, and because neither figure is declared either
153 
  # way. The Gemini figures were written to make the system run and were not
154 
  # read off any price page.
142 
  # Declared rates are the provider's published table, entered here so a call
143 
  # can be billed (METER-001). A time-bounded `promotion` replaces that table
144 
  # until its exclusive UTC cutoff, then the regular table applies without
145 
  # another deployment.
155 146 
  #
156 147 
  # A model with no `pricing` key records no estimated cost at all — not a zero
157 148 
  # — and its usage records stamp `pricing_id: "unpriced"`. No entry below is in

@@ -251,15 +242,24 @@ config :openagents,

251 242 
      provider_model: "google/gemini-3.7-flash",
252 243 
      context_window: 1_048_576,
253 244 
      max_output: 65_536,
254 
      # Placeholder: the operator must set real provider rates and flip `source`
255 
      # to `:declared` before anything bills from this. The cached-input rate is
256 
      # optional and should be omitted if the provider does not offer one.
245 
      # Google's published Standard paid-tier table for Gemini 3.7 Flash
246 
      # (https://ai.google.dev/gemini-api/docs/pricing). List rates apply from
247 
      # 2027-01-01; the introductory promotion is half those rates through the
248 
      # exclusive UTC cutoff, the same shape as GLM's promotion.
257 249 
      pricing: %{
258 
        id: "placeholder.gemini-3.7-flash.v1",
259 
        source: :placeholder,
260 
        input_per_million_tokens: 1_250_000,
261 
        output_per_million_tokens: 10_000_000,
262 
        cached_input_per_million_tokens: 100_000
250 
        id: "declared.gemini-3.7-flash.v1",
251 
        source: :declared,
252 
        input_per_million_tokens: 1_500_000,
253 
        output_per_million_tokens: 7_500_000,
254 
        cached_input_per_million_tokens: 150_000,
255 
        promotion: %{
256 
          id: "declared.gemini-3.7-flash.intro-through-2026-12-31.v1",
257 
          source: :declared,
258 
          ends_at: ~U[2027-01-01 00:00:00Z],
259 
          input_per_million_tokens: 750_000,
260 
          output_per_million_tokens: 3_750_000,
261 
          cached_input_per_million_tokens: 75_000
262 
        }
263 263 
      }
264 264 
    },
265 265 
    %{
lib/openagents_web/live/model_catalog_live.ex modified +2 -1

@@ -21,7 +21,8 @@ defmodule OpenAgentsWeb.ModelCatalogLive do

21 21 
  different fact from the call having cost nothing. And a rate this deployment
22 22 
  wrote to make itself run is labelled `provisional`, so nobody reads a working
23 23 
  figure as a price. The Coder Free lane declares zero rates because its router
24 
  only selects free models; the other listed rates remain provisional.
24 
  only selects free models; GLM 5.3 Flash and Gemini 3.7 Flash declare the
25 
  provider's published rates, including any active promotion.
25 26
26 27 
  Availability is `Models.availability/1`, the same word the endpoint
27 28 
  publishes, refreshed on a slow tick: a page about what a lane costs that
test/openagents/inference/models_test.exs modified +7 -3

@@ -152,9 +152,13 @@ defmodule OpenAgents.Inference.ModelsTest do

152 152 
      gemini = Enum.find(Models.catalog(), &(&1["id"] == "gemini-3.7-flash"))
153 153
154 154 
      assert %{"pricing" => pricing} = gemini
155 
      assert pricing["input_per_million_tokens"] == 1_250_000
156 
      assert pricing["output_per_million_tokens"] == 10_000_000
157 
      assert pricing["cached_input_per_million_tokens"] == 100_000
155 
      assert pricing["id"] == "declared.gemini-3.7-flash.intro-through-2026-12-31.v1"
156 
      assert pricing["basis"] == "declared"
157 
      assert pricing["input_per_million_tokens"] == 750_000
158 
      assert pricing["output_per_million_tokens"] == 3_750_000
159 
      assert pricing["cached_input_per_million_tokens"] == 75_000
160 
      assert gemini["pricing_promotion"]["ends_at"] == "2027-01-01T00:00:00Z"
161 
      assert gemini["pricing_promotion"]["active"]
158 162 
    end
159 163
160 164 
    test "the default publishes its currently effective rate table" do
test/openagents/inference/pricing_test.exs modified +40 -15

@@ -43,7 +43,8 @@ defmodule OpenAgents.Inference.PricingTest do

43 43 
    end
44 44
45 45 
    test "a lane with rates publishes the table those rates came from" do
46 
      assert Pricing.pricing_id_for("gemini-3.7-flash") == "placeholder.gemini-3.7-flash.v1"
46 
      assert Pricing.pricing_id_for("gemini-3.7-flash") ==
47 
               "declared.gemini-3.7-flash.intro-through-2026-12-31.v1"
47 48 
    end
48 49
49 50 
    test "rates without a table name are unattributed rather than silently accepted" do

@@ -88,6 +89,30 @@ defmodule OpenAgents.Inference.PricingTest do

88 89 
    end
89 90 
  end
90 91
92 
  describe "the Gemini 3.7 Flash published rates" do
93 
    test "uses intro rates before the cutoff and list rates at the cutoff" do
94 
      {:ok, gemini} = OpenAgents.Inference.Models.fetch("gemini-3.7-flash")
95
96 
      introductory = Pricing.effective_pricing(gemini, ~U[2026-12-31 23:59:59.999999Z])
97 
      list = Pricing.effective_pricing(gemini, ~U[2027-01-01 00:00:00Z])
98
99 
      assert introductory.id == "declared.gemini-3.7-flash.intro-through-2026-12-31.v1"
100 
      assert introductory.source == :declared
101 
      assert introductory.input_per_million_tokens == 750_000
102 
      assert introductory.output_per_million_tokens == 3_750_000
103 
      assert introductory.cached_input_per_million_tokens == 75_000
104 
      assert Pricing.promotion_active?(gemini, ~U[2026-12-31 23:59:59Z])
105
106 
      assert list.id == "declared.gemini-3.7-flash.v1"
107 
      assert list.source == :declared
108 
      assert list.input_per_million_tokens == 1_500_000
109 
      assert list.output_per_million_tokens == 7_500_000
110 
      assert list.cached_input_per_million_tokens == 150_000
111 
      refute Pricing.promotion_active?(gemini, ~U[2027-01-01 00:00:00Z])
112 
      assert Pricing.promotion_ends_at(gemini) == ~U[2027-01-01 00:00:00Z]
113 
    end
114 
  end
115
91 116 
  describe "pricing one usage record" do
92 117 
    test "an unpriced model writes no cost key — not a zero" do
93 118 
      priced =

@@ -104,9 +129,10 @@ defmodule OpenAgents.Inference.PricingTest do

104 129 
      priced =
105 130 
        Pricing.price(%{"input_tokens" => 1_000_000, "output_tokens" => 0}, "gemini-3.7-flash")
106 131
107 
      assert priced["estimated_cost_microusd"] == 1_250_000
108 
      assert priced["pricing_id"] == "placeholder.gemini-3.7-flash.v1"
109 
      assert Pricing.cost(priced) == 1_250_000
132 
      # Test `pricing_now` is 2026-09-01, inside the introductory promotion.
133 
      assert priced["estimated_cost_microusd"] == 750_000
134 
      assert priced["pricing_id"] == "declared.gemini-3.7-flash.intro-through-2026-12-31.v1"
135 
      assert Pricing.cost(priced) == 750_000
110 136 
    end
111 137
112 138 
    test "cached reads are charged at the cached rate and cache writes at the input rate" do

@@ -121,19 +147,19 @@ defmodule OpenAgents.Inference.PricingTest do

121 147 
          "gemini-3.7-flash"
122 148 
        )
123 149
124 
      # 600k uncached input + 100k cache write at $1.25/M, 400k cached read at
125 
      # $0.10/M: 700_000 * 1.25 + 400_000 * 0.10, in microUSD.
126 
      assert priced["estimated_cost_microusd"] == 875_000 + 40_000
150 
      # 600k uncached input + 100k cache write at $0.75/M, 400k cached read at
151 
      # $0.075/M: 700_000 * 0.75 + 400_000 * 0.075, in microUSD.
152 
      assert priced["estimated_cost_microusd"] == 525_000 + 30_000
127 153 
    end
128 154 
  end
129 155
130 156 
  describe "reading a stored record back" do
131 
    test "a placeholder-priced record carries a cost but is not billable" do
157 
    test "a declared Gemini record carries a cost and is billable" do
132 158 
      priced = Pricing.price(%{"input_tokens" => 1_000_000}, "gemini-3.7-flash")
133 159
134 160 
      assert Pricing.priced?(priced)
135 
      assert Pricing.usage_basis(priced) == "provisional"
136 
      refute Pricing.billable?(priced)
161 
      assert Pricing.usage_basis(priced) == "declared"
162 
      assert Pricing.billable?(priced)
137 163 
    end
138 164
139 165 
    test "an unpriced record is never billable" do

@@ -176,8 +202,7 @@ defmodule OpenAgents.Inference.PricingTest do

176 202
177 203 
      glm = Enum.find(OpenAgents.Inference.Models.catalog(), &(&1["id"] == "glm-5.3-flash"))
178 204
179 
      assert Enum.count(bases, &(&1 == "declared")) >= 2
180 
      assert "provisional" in bases
205 
      assert Enum.all?(bases, &(&1 == "declared"))
181 206 
      assert free_router["pricing"]["input_per_million_tokens"] == 0
182 207 
      assert free_router["pricing"]["output_per_million_tokens"] == 0
183 208 
      assert glm["pricing"]["input_per_million_tokens"] == 150_000

@@ -196,9 +221,9 @@ defmodule OpenAgents.Inference.PricingTest do

196 221 
        OpenAgents.Inference.Models.catalog()
197 222 
        |> Enum.find(&(&1["id"] == "gemini-3.7-flash"))
198 223
199 
      assert gemini["pricing"]["id"] == "placeholder.gemini-3.7-flash.v1"
200 
      assert gemini["pricing"]["basis"] == "provisional"
201 
      assert gemini["pricing_basis"] == "provisional"
224 
      assert gemini["pricing"]["id"] == "declared.gemini-3.7-flash.intro-through-2026-12-31.v1"
225 
      assert gemini["pricing"]["basis"] == "declared"
226 
      assert gemini["pricing_basis"] == "declared"
202 227 
    end
203 228 
  end
204 229 
end
test/openagents_web/controllers/inference_proxy_fallback_test.exs modified +4 -1

@@ -261,7 +261,10 @@ defmodule OpenAgentsWeb.InferenceProxyFallbackTest do

261 261
262 262 
      metered = Repo.get(Grant, grant.id)
263 263 
      assert metered.usage["served_model"] == @gemini
264 
      assert metered.usage["pricing_id"] == "placeholder.gemini-3.7-flash.v1"
264
265 
      assert metered.usage["pricing_id"] ==
266 
               "declared.gemini-3.7-flash.intro-through-2026-12-31.v1"
267
265 268 
      assert Pricing.cost(metered.usage) > 0
266 269 
      assert Health.status(@gemini) == {:healthy, nil}
267 270 
    end
test/openagents_web/controllers/model_catalog_controller_test.exs modified +6 -6

@@ -110,9 +110,9 @@ defmodule OpenAgentsWeb.ModelCatalogControllerTest do

110 110
111 111 
      assert %{
112 112 
               "pricing" => %{
113 
                 "input_per_million_tokens" => 1_250_000,
114 
                 "output_per_million_tokens" => 10_000_000,
115 
                 "cached_input_per_million_tokens" => 100_000
113 
                 "input_per_million_tokens" => 750_000,
114 
                 "output_per_million_tokens" => 3_750_000,
115 
                 "cached_input_per_million_tokens" => 75_000
116 116 
               }
117 117 
             } = gemini
118 118 
    end

@@ -151,9 +151,9 @@ defmodule OpenAgentsWeb.ModelCatalogControllerTest do

151 151 
               "unpriced"
152 152
153 153 
      gemini = Enum.find(body["models"], &(&1["id"] == "gemini-3.7-flash"))
154 
      assert gemini["pricing_basis"] == "provisional"
155 
      assert gemini["pricing"]["basis"] == "provisional"
156 
      assert gemini["pricing"]["id"] == "placeholder.gemini-3.7-flash.v1"
154 
      assert gemini["pricing_basis"] == "declared"
155 
      assert gemini["pricing"]["basis"] == "declared"
156 
      assert gemini["pricing"]["id"] == "declared.gemini-3.7-flash.intro-through-2026-12-31.v1"
157 157 
    end
158 158 
  end
159 159 
end
test/openagents_web/live/model_catalog_live_test.exs modified +5 -4

@@ -55,12 +55,13 @@ defmodule OpenAgentsWeb.ModelCatalogLiveTest do

55 55 
  test "a lane with rates publishes them and names the table they came from", %{conn: conn} do
56 56 
    {:ok, view, _html} = live(reader(conn, "model-catalog-priced"), ~p"/models")
57 57
58 
    # 1_250_000 microUSD per million tokens is $1.25 per million tokens.
59 
    assert view |> element(~s([id="model-input-rate-gemini-3.7-flash"])) |> render() =~ "$1.25"
60 
    assert view |> element(~s([id="model-basis-gemini-3.7-flash"])) |> render() =~ "provisional"
58 
    # 750_000 microUSD per million tokens is $0.75 per million tokens, the
59 
    # introductory Standard paid-tier rate still in force at test `pricing_now`.
60 
    assert view |> element(~s([id="model-input-rate-gemini-3.7-flash"])) |> render() =~ "$0.75"
61 
    assert view |> element(~s([id="model-basis-gemini-3.7-flash"])) |> render() =~ "declared"
61 62
62 63 
    assert view |> element(~s([id="model-pricing-id-gemini-3.7-flash"])) |> render() =~
63 
             "placeholder.gemini-3.7-flash.v1"
64 
             "declared.gemini-3.7-flash.intro-through-2026-12-31.v1"
64 65 
  end
65 66
66 67 
  test "the declared free router is billable at zero", %{conn: conn} do

This page updates live while a promote is in flight · changelog