systemprompt-models 0.53.0

# Vertex AI rate card: the prices we bill, the allowlist of Vertex models the
# gateway will serve, and the lifecycle Google's documentation gives each one.
#
# Why one file for all three jobs: Vertex's publisher listing carries no
# modality field, so `gemini-3.8-flash` (chat), `gemini-embedding-001`
# (embeddings) and `gemini-2.5-flash-tts` (speech) are indistinguishable by
# shape, and it carries no retirement either -- Vertex lists a model right up
# to the day it is switched off. Discovery therefore cannot decide what is
# chat-capable or still supported on its own. An entry here is that decision;
# it has to carry a price before the gateway will dispatch to it (boot refuses
# an unpriced route), and it has to carry the documented lifecycle before
# discovery will publish it. A model absent from this file is reported as
# discovered-but-unpriced and never published; a model past (or within 30
# days of) `retires_on`, or `preview` without `allow_preview`, is reported as
# retiring and never published. No model is ever called to find this out.
#
# `upstream` is the model as Vertex names it (`{publisher}/{model}` from
# `publishers/{publisher}/models/{model}`); `upstream_model` is what our wire
# puts on the request -- the bare name for the gemini wire, the publisher-
# qualified name for the MaaS openai-chat surface.
#
# `launch_stage`, `released`, `retires_on`, `price_until` and every price are
# read from the `docs` page named on the entry (all read 2026-09-11). Gemini
# 3.x is served from the `global` location only; a regional Vertex endpoint
# 404s on every one of them.
#
# `allow_preview` publishes a model whose stage is not GA. It is off by
# default: a preview model can change or vanish under a running deployment.
entries:
# -- provider `vertex` (wire: gemini) -----------------------------------------
# Cache writes are billed as storage per hour, not per token, hence an explicit
# 0.0 on every Gemini entry.
# Why intro pricing: Google lists 0.75/3.75 as introductory to 2026-12-31, then
# 1.50/7.50. `price_until` is the step-up date; the scheduler warns 60 days
# ahead so the card is re-read before the standard rate applies.
- upstream: google/gemini-3.8-flash
  provider: vertex
  id: vertex-gemini-3.8-flash
  upstream_model: gemini-3.8-flash
  launch_stage: ga
  released: 2026-09-02
  price_until: 2026-12-31
  docs: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/3-8-flash
  pricing:
    input_per_million: 0.75
    output_per_million: 3.75
    cache_read_per_million: 0.075
    cache_write_per_million: 0.0
    per_image_cents: null
  capabilities:
    vision: true
    streaming: true
    tools: true
    structured_output: true
    system_prompts: true
    reasoning: true
  limits:
    context_window: 1000000
    max_output_tokens: 65536
    max_thinking_budget: 24576
- upstream: google/gemini-3.7-flash
  provider: vertex
  id: vertex-gemini-3.7-flash
  upstream_model: gemini-3.7-flash
  launch_stage: ga
  released: 2026-08-13
  price_until: 2026-12-31
  docs: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/3-7-flash
  pricing:
    input_per_million: 0.75
    output_per_million: 3.75
    cache_read_per_million: 0.075
    cache_write_per_million: 0.0
    per_image_cents: null
  capabilities:
    vision: true
    streaming: true
    tools: true
    structured_output: true
    system_prompts: true
    reasoning: true
  limits:
    context_window: 1000000
    max_output_tokens: 65536
    max_thinking_budget: 24576
- upstream: google/gemini-3.6-flash
  provider: vertex
  id: vertex-gemini-3.6-flash
  upstream_model: gemini-3.6-flash
  launch_stage: ga
  released: 2026-07-21
  price_until: 2026-12-31
  docs: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/3-6-flash
  pricing:
    input_per_million: 0.75
    output_per_million: 3.75
    cache_read_per_million: 0.075
    cache_write_per_million: 0.0
    per_image_cents: null
  capabilities:
    vision: true
    streaming: true
    tools: true
    structured_output: true
    system_prompts: true
    reasoning: true
  limits:
    context_window: 1000000
    max_output_tokens: 65536
    max_thinking_budget: 24576
- upstream: google/gemini-3.5-flash
  provider: vertex
  id: vertex-gemini-3.5-flash
  upstream_model: gemini-3.5-flash
  launch_stage: ga
  released: 2026-05-19
  retires_on: 2027-05-19
  docs: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/3-5-flash
  pricing:
    input_per_million: 1.5
    output_per_million: 9.0
    cache_read_per_million: 0.15
    cache_write_per_million: 0.0
    per_image_cents: null
  capabilities:
    vision: true
    streaming: true
    tools: true
    structured_output: true
    system_prompts: true
    reasoning: true
  limits:
    context_window: 1000000
    max_output_tokens: 65536
    max_thinking_budget: 24576
- upstream: google/gemini-3.5-flash-lite
  provider: vertex
  id: vertex-gemini-3.5-flash-lite
  upstream_model: gemini-3.5-flash-lite
  launch_stage: ga
  released: 2026-07-21
  retires_on: 2027-07-21
  docs: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/3-5-flash-lite
  pricing:
    input_per_million: 0.3
    output_per_million: 2.5
    cache_read_per_million: 0.03
    cache_write_per_million: 0.0
    per_image_cents: null
  capabilities:
    vision: true
    streaming: true
    tools: true
    structured_output: true
    system_prompts: true
    reasoning: true
  limits:
    context_window: 1000000
    max_output_tokens: 65536
    max_thinking_budget: 24576
- upstream: google/gemini-3.1-flash-lite
  provider: vertex
  id: vertex-gemini-3.1-flash-lite
  upstream_model: gemini-3.1-flash-lite
  launch_stage: ga
  released: 2026-05-07
  retires_on: 2027-05-07
  docs: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/3-1-flash-lite
  pricing:
    input_per_million: 0.25
    output_per_million: 1.5
    cache_read_per_million: 0.025
    cache_write_per_million: 0.0
    per_image_cents: null
  capabilities:
    vision: true
    streaming: true
    tools: true
    structured_output: true
    system_prompts: true
    reasoning: true
  limits:
    context_window: 1000000
    max_output_tokens: 65536
    max_thinking_budget: 24576
# Why allow_preview: 3.1 Pro has no GA successor yet, so it is the only Pro
# tier Google serves; it has been priced and documented since February.
- upstream: google/gemini-3.1-pro-preview
  provider: vertex
  id: vertex-gemini-3.1-pro-preview
  upstream_model: gemini-3.1-pro-preview
  allow_preview: true
  launch_stage: preview
  released: 2026-02-19
  docs: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/3-1-pro
  pricing:
    input_per_million: 2.0
    output_per_million: 12.0
    cache_read_per_million: 0.2
    cache_write_per_million: 0.0
    per_image_cents: null
  capabilities:
    vision: true
    streaming: true
    tools: true
    structured_output: true
    system_prompts: true
    reasoning: true
  limits:
    context_window: 1000000
    max_output_tokens: 65536
    max_thinking_budget: 32768
# The 2.5 family retires 2026-10-20 (models/gemini/lifecycle). Kept priced so
# an explicit catalog declaration still bills correctly; discovery stops
# publishing them 30 days before the date.
- upstream: google/gemini-2.5-pro
  provider: vertex
  id: vertex-gemini-2.5-pro
  upstream_model: gemini-2.5-pro
  launch_stage: ga
  released: 2025-06-17
  retires_on: 2026-10-20
  docs: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/2-5-pro
  pricing:
    input_per_million: 1.25
    output_per_million: 10.0
    cache_read_per_million: 0.31
    cache_write_per_million: 0.0
    per_image_cents: null
  capabilities:
    vision: true
    streaming: true
    tools: true
    structured_output: true
    system_prompts: true
    reasoning: true
  limits:
    context_window: 1000000
    max_output_tokens: 65536
    max_thinking_budget: 32768
- upstream: google/gemini-2.5-flash
  provider: vertex
  id: vertex-gemini-2.5-flash
  upstream_model: gemini-2.5-flash
  launch_stage: ga
  released: 2025-06-17
  retires_on: 2026-10-20
  docs: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/2-5-flash
  pricing:
    input_per_million: 0.3
    output_per_million: 2.5
    cache_read_per_million: 0.075
    cache_write_per_million: 0.0
    per_image_cents: null
  capabilities:
    vision: true
    streaming: true
    tools: true
    structured_output: true
    system_prompts: true
    reasoning: true
  limits:
    context_window: 1000000
    max_output_tokens: 65536
    max_thinking_budget: 24576
- upstream: google/gemini-2.5-flash-lite
  provider: vertex
  id: vertex-gemini-2.5-flash-lite
  upstream_model: gemini-2.5-flash-lite
  launch_stage: ga
  released: 2025-07-22
  retires_on: 2026-10-20
  docs: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/2-5-flash-lite
  pricing:
    input_per_million: 0.1
    output_per_million: 0.4
    cache_read_per_million: 0.025
    cache_write_per_million: 0.0
    per_image_cents: null
  capabilities:
    vision: true
    streaming: true
    tools: true
    structured_output: true
    system_prompts: true
    reasoning: true
  limits:
    context_window: 1000000
    max_output_tokens: 65536
    max_thinking_budget: 24576
# -- provider `vertex-maas` (wire: openai-chat) -------------------------------
# Why: cache rates are what each model page publishes; a page with no cache
# tier gets an explicit 0.0 -- declared free, not forgotten.
#
# `max_thinking_budget` marks a model as reasoning for the openai-chat wire,
# which then sends the model's own output cap instead of the caller's
# max_tokens so hidden reasoning cannot starve the visible answer.
- upstream: qwen/qwen3-coder-480b-a35b-instruct-maas
  provider: vertex-maas
  id: qwen.qwen3-coder-480b
  upstream_model: qwen/qwen3-coder-480b-a35b-instruct-maas
  launch_stage: ga
  released: 2025-08-13
  docs: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/maas/qwen/qwen3-coder
  pricing:
    input_per_million: 0.22
    output_per_million: 1.8
    cache_read_per_million: 0.022
    cache_write_per_million: 0.0
    per_image_cents: null
  capabilities:
    streaming: true
    tools: true
    structured_output: true
    system_prompts: true
  limits:
    context_window: 262144
    max_output_tokens: 65536
# Why allow_preview: the documentation calls glm-5 GA but the Vertex listing
# still marks it EXPERIMENTAL; without the opt-in the only GA GLM would be
# withheld on the listing's word against the docs'.
- upstream: zai-org/glm-5-maas
  provider: vertex-maas
  id: zai.glm-5
  upstream_model: zai-org/glm-5-maas
  allow_preview: true
  launch_stage: ga
  released: 2026-02-10
  docs: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/maas/zaiorg/glm-5
  pricing:
    input_per_million: 1.0
    output_per_million: 3.2
    cache_read_per_million: 0.1
    cache_write_per_million: 0.0
    per_image_cents: null
  capabilities:
    streaming: true
    tools: true
    structured_output: true
    system_prompts: true
  limits:
    context_window: 200000
    max_output_tokens: 32768
    max_thinking_budget: 16384
# Why allow_preview: 5.2 is the current GLM and is served beside 5 so a
# developer can choose; it is priced and documented, preview or not.
- upstream: zai-org/glm-5.2-maas
  provider: vertex-maas
  id: zai.glm-5.2
  upstream_model: zai-org/glm-5.2-maas
  allow_preview: true
  launch_stage: preview
  released: 2026-08-24
  docs: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/maas/zaiorg/glm-52
  pricing:
    input_per_million: 1.4
    output_per_million: 4.4
    cache_read_per_million: 0.14
    cache_write_per_million: 0.0
    per_image_cents: null
  capabilities:
    streaming: true
    tools: true
    structured_output: true
    system_prompts: true
  limits:
    context_window: 200000
    max_output_tokens: 32768
    max_thinking_budget: 16384
- upstream: openai/gpt-oss-120b-maas
  provider: vertex-maas
  id: openai.gpt-oss-120b
  upstream_model: openai/gpt-oss-120b-maas
  launch_stage: ga
  released: 2025-08-13
  docs: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/maas/openai/gpt-oss-120b
  pricing:
    input_per_million: 0.09
    output_per_million: 0.36
    cache_read_per_million: 0.0
    cache_write_per_million: 0.0
    per_image_cents: null
  capabilities:
    streaming: true
    tools: true
    structured_output: true
    system_prompts: true
  limits:
    context_window: 128000
    max_output_tokens: 32768
    max_thinking_budget: 16384
# Deprecated 2026-07-21, retiring 2026-10-21 (models/deprecations/partner-models).
# Kept priced so an explicit declaration bills correctly until the date;
# discovery withholds them from 2026-09-21. Delete after retirement. Their
# model pages are gone, so `released` is omitted rather than guessed.
- upstream: qwen/qwen3-next-80b-a3b-instruct-maas
  provider: vertex-maas
  id: qwen.qwen3-next-instruct
  upstream_model: qwen/qwen3-next-80b-a3b-instruct-maas
  launch_stage: ga
  retires_on: 2026-10-21
  docs: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/deprecations/partner-models
  pricing:
    input_per_million: 0.15
    output_per_million: 1.2
    cache_read_per_million: 0.0
    cache_write_per_million: 0.0
    per_image_cents: null
  capabilities:
    streaming: true
    tools: true
    structured_output: true
    system_prompts: true
  limits:
    context_window: 262144
    max_output_tokens: 32768
- upstream: qwen/qwen3-next-80b-a3b-thinking-maas
  provider: vertex-maas
  id: qwen.qwen3-next-thinking
  upstream_model: qwen/qwen3-next-80b-a3b-thinking-maas
  launch_stage: ga
  retires_on: 2026-10-21
  docs: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/deprecations/partner-models
  pricing:
    input_per_million: 0.15
    output_per_million: 1.2
    cache_read_per_million: 0.0
    cache_write_per_million: 0.0
    per_image_cents: null
  capabilities:
    streaming: true
    tools: true
    structured_output: true
    system_prompts: true
    reasoning: true
  limits:
    context_window: 262144
    max_output_tokens: 32768
    max_thinking_budget: 16384
- upstream: qwen/qwen3-235b-a22b-instruct-2507-maas
  provider: vertex-maas
  id: qwen.qwen3-235b
  upstream_model: qwen/qwen3-235b-a22b-instruct-2507-maas
  launch_stage: ga
  retires_on: 2026-10-21
  docs: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/deprecations/partner-models
  pricing:
    input_per_million: 0.25
    output_per_million: 1.0
    cache_read_per_million: 0.0
    cache_write_per_million: 0.0
    per_image_cents: null
  capabilities:
    streaming: true
    tools: true
    structured_output: true
    system_prompts: true
  limits:
    context_window: 262144
    max_output_tokens: 16384
- upstream: moonshotai/kimi-k2-thinking-maas
  provider: vertex-maas
  id: moonshotai.kimi-k2-thinking
  upstream_model: moonshotai/kimi-k2-thinking-maas
  launch_stage: ga
  retires_on: 2026-10-21
  docs: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/deprecations/partner-models
  pricing:
    input_per_million: 0.6
    output_per_million: 2.5
    cache_read_per_million: 0.0
    cache_write_per_million: 0.0
    per_image_cents: null
  capabilities:
    streaming: true
    tools: true
    structured_output: true
    system_prompts: true
    reasoning: true
  limits:
    context_window: 262144
    max_output_tokens: 32768
    max_thinking_budget: 16384
- upstream: deepseek-ai/deepseek-v3.2-maas
  provider: vertex-maas
  id: deepseek.v3.2
  upstream_model: deepseek-ai/deepseek-v3.2-maas
  launch_stage: ga
  retires_on: 2026-10-21
  docs: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/deprecations/partner-models
  pricing:
    input_per_million: 0.56
    output_per_million: 1.68
    cache_read_per_million: 0.0
    cache_write_per_million: 0.0
    per_image_cents: null
  capabilities:
    streaming: true
    tools: true
    structured_output: true
    system_prompts: true
  limits:
    context_window: 128000
    max_output_tokens: 32768
    max_thinking_budget: 16384
- upstream: openai/gpt-oss-20b-maas
  provider: vertex-maas
  id: openai.gpt-oss-20b
  upstream_model: openai/gpt-oss-20b-maas
  launch_stage: ga
  released: 2025-08-13
  retires_on: 2026-10-21
  docs: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/deprecations/partner-models
  pricing:
    input_per_million: 0.07
    output_per_million: 0.3
    cache_read_per_million: 0.0
    cache_write_per_million: 0.0
    per_image_cents: null
  capabilities:
    streaming: true
    tools: true
    structured_output: true
    system_prompts: true
  limits:
    context_window: 128000
    max_output_tokens: 32768
    max_thinking_budget: 16384