Files
zeekay bd98d0f7f2 feat(ai/models): enable + capability-declare the full do-ai premium chat lineup
Expose the top DigitalOcean GenAI chat models to customers on /v1/models — the
Claude family (opus-4.8/4.7, sonnet-5, sonnet-4.6, haiku-4.5, fable-5), the
OpenAI family (gpt-5.6 luna/sol/terra, gpt-5.5, gpt-5.4, gpt-4o, gpt-4o-mini,
o3), and the strong open models (deepseek-v4-pro, llama-4-maverick, qwen3.5-397b,
glm-5.2, kimi-k2.6). Each was hidden or missing in conf/models.yaml (the runtime
source of truth), so /v1/models listed ~35 models and none of the premium chat
SKUs. Now 55 are listed. All route to the do-ai provider, so they meter under
do-ai unchanged.

Every model + capability verified LIVE against inference.do-ai.run:
- context_window + max_output_tokens from the DO catalog (authoritative).
- supports_vision / supports_tools set ONLY where the probe confirmed it:
  * gpt-5.6 (all three variants) REJECTS tool calls (HTTP 400) -> tools off.
  * claude-sonnet-5, deepseek-v4-pro, glm-5.2, kimi-k2.6, qwen3.5 REJECT image
    input (HTTP 400) -> vision off.
  * claude-sonnet-4-6 re-pointed to its REAL anthropic-claude-4.6-sonnet upstream
    (now served on this account; previously degraded to 4.5 while 4.6 403'd) and
    the stale lineup-test expectation + comments updated to match.

Schema: add omitempty max_output_tokens / supports_vision / supports_tools to
modelInfo (the /v1/models shape) + ModelDef (YAML) + modelRoute, wired through
ListModels and the static fallback. Additive: a legacy consumer is unaffected,
and a "no" is expressed by absence, never a fabricated false.

Router: append the newly-enabled models to the conf `prefer` lists as candidates
(zen champions stay first, so the default/cost is unchanged; explore floor and
the learned engine can now reach them). Only capability-matched ids per task —
confirmed-vision-only in `vision`, no gpt-5.6 in tool-heavy code/reasoning.

Drive-by: drop a vet-flagged self-assignment in object/finetune_k8s.go.

Verified: go build ./... clean; go vet ./controllers/ ./object/ clean;
go test ./controllers/ ./object/ -run 'Model|Provider|Catalog|Route' green, plus
new TestDOAIEnabledModelsVisibleAndCapable / TestModelInfoCapabilityFieldsOmitEmpty.

Follow-up: sync the prod cloud-api-models ConfigMap
(universe/infra/k8s/cloud/models.yaml) — bootstrap reads conf/models.yaml, but
prod mounts that ConfigMap over it.
2026-07-18 23:29:45 -07:00

1231 lines
39 KiB
YAML

# Hanzo Cloud Model Configuration
# This file defines model routing, pricing, and identity prompts.
# In production with live_mode: true, pricing is refreshed from pricing.hanzo.ai.
version: 1
services:
pricing_url: "https://pricing.hanzo.ai" # Override: PRICING_SERVICE_URL env var
cache:
pricing_ttl: "6h"
features:
live_mode: false # Set true in production ConfigMap
premium_gate: true
# ── Auto-routing (virtual `auto` / `zen-router` model) ────────────────────────
# Opt-in. When enabled, a chat request for model "auto" (alias "zen-router") is
# classified into a coarse task and mapped to the first SERVABLE model in the
# matching prefer list, BEFORE provider/pricing/billing resolution — so the
# request is billed and reported as the model that served it (see the
# X-Routed-Model response header). Disabled by default: "auto" is then just an
# unknown model id.
#
# Strategy: if `endpoint` is set, POST {endpoint}/route to a zen-router served by
# hanzo-engine (learned routing); on ANY error (or when endpoint is empty) fall
# back to the built-in pure-Go heuristic — a faithful port of hanzo-router's
# classifier — so auto works with ZERO extra infrastructure. Env overrides:
# ROUTER_ENABLED=true, ROUTER_ENDPOINT=<url>.
router:
enabled: true # global auto-routing default: ON for everyone (AutoRouting); org/"*" OrgSettings rows override
# default_session_routing has no conf flag — its conf-level default is disabled;
# set it per-org or platform-wide via an OrgSettings row (owner "*").
endpoint: "" # e.g. http://engine.hanzo.svc:3000 (zen-router via hanzo-engine)
cost_ceiling: 0.0 # advisory per-1k cost cap, forwarded to the engine as SLO
prefer: # task tag → ordered model ids; first servable wins ("default" = catch-all)
# zen champions stay FIRST (unchanged default/cost); the newly-enabled do-ai
# premium models are appended as CANDIDATES — reachable by the learned engine
# and the exploration floor, and as fallbacks when a champion is unservable.
# Only models VERIFIED to have the task's capability are listed (e.g. only
# confirmed-vision ids go in `vision`; gpt-5.6 has no tools so it stays out of
# tool-heavy code/reasoning; sonnet-5/deepseek-v4-pro/glm-5.2 reject images so
# they are absent from `vision`).
code: [zen4-coder, zen5-coder, qwen3-coder, gpt-5.3-codex, claude-opus-4-8, claude-sonnet-4-6, gpt-5.4, deepseek-v4-pro]
reasoning: [zen4-thinking, deepseek-reasoner, zen4-ultra, o3, gpt-5.5, claude-opus-4-8, deepseek-v4-pro]
math: [zen4-ultra, deepseek-reasoner, zen4-thinking, o3, gpt-5.5, deepseek-v4-pro]
creative: [zen4, zen4-max, claude-sonnet-4-5, claude-opus-4-8, claude-fable-5, gpt-5.5]
vision: [zen5-omni, zen3-omni, zen3-vl, claude-sonnet-4-6, gpt-4o, gpt-5.4, llama-4-maverick, nemotron-nano-vl]
long_context: [zen4-max, gpt-5.4, claude-sonnet-4-5, gpt-5.5, gpt-5.6-luna, claude-opus-4-8, deepseek-v4-pro]
cheap_chat: [zen4-mini, zen5-mini, gpt-4o-mini, claude-haiku-4-5]
default: [zen4, zen5, gpt-4o, gpt-4o-mini, kimi-k2.6, glm-5.2]
default_pricing:
input_per_million: 1.00
output_per_million: 4.00
# How long we hold a request open for a provider that is merely busy, instead
# of handing its 429 to the client. A 429 means "ask again shortly" — and we
# can do that far better than a client can: it cannot pick another provider,
# and if every client retries at once they arrive together and overload the
# provider again. Tunable here (and from admin) without a rebuild.
#
# attempts: total tries of ONE provider (1 = no same-provider retry).
# base_backoff: first wait; doubles each attempt, jittered to avoid a herd.
# max_backoff: ceiling on any single wait. A provider still refusing after a
# few seconds is not blipping — the cure is a different provider
# (the fallbacks: chain), not a longer sleep.
retry:
attempts: 3
base_backoff: 500ms
max_backoff: 4s
models:
# ── `best` — auto-routing "best available coding model" (quality-ranked) ─
# A virtual policy model: it resolves to the FIRST SERVABLE model in the
# chain below, cascading to the next rank when one is rate-limited, out of
# credit, agreement-gated, or down (402/403/429/quota/credit/5xx/unavailable)
# via the deployed failover machinery (controllers/failover.go). This is the
# DRY "one way": `best` is just a model whose fallback chain IS the quality
# reordering the list. `hanzo code` defaults to this model. A SuperAdmin can
# override the ranking at runtime WITHOUT a restart via a DB model_route for
# `best` (owner "built-in"), which resolves before this file and hot-reloads
# on write (60s cache); this file is the durable default. Future `cheap`/
# `fast` policy models are the same mechanism with a different order.
# NOTE: prod serves the catalog from the cloud-api-models ConfigMap
# (universe/infra/k8s/cloud/models.yaml) — keep `best` in sync there.
best:
context_window: 262144
provider: do-ai
upstream: glm-5.2
owned_by: hanzo
fallbacks:
- {provider: do-ai, upstream: deepseek-v4-pro, context_window: 1000000}
- {provider: do-ai, upstream: kimi-k2.6}
- {provider: do-ai, upstream: deepseek-4-flash}
- {provider: do-ai, upstream: qwen3.5-397b-a17b}
pricing: {input: 0.60, output: 2.20}
# ── DO-AI models (non-premium, included in free credit) ────────────────
# OpenAI via DO-AI
gpt-4o:
provider: do-ai
upstream: openai-gpt-4o
context_window: 128000
max_output_tokens: 16384
vision: true # verified live: answered image color
tools: true # verified live: emitted tool_calls
fallbacks:
- provider: openai-direct
upstream: gpt-4o
pricing: {input: 2.50, output: 10.00}
gpt-4o-mini:
provider: do-ai
upstream: openai-gpt-4o-mini
context_window: 128000
max_output_tokens: 16384
vision: true # verified live
tools: true # verified live
fallbacks:
- provider: openai-direct
upstream: gpt-4o-mini
pricing: {input: 0.15, output: 0.60}
gpt-4.1:
provider: do-ai
upstream: openai-gpt-4.1
fallbacks:
- provider: openai-direct
upstream: gpt-4.1
pricing: {input: 2.00, output: 8.00}
hidden: true
gpt-5:
provider: do-ai
upstream: openai-gpt-5
fallbacks:
- provider: openai-direct
upstream: gpt-5
pricing: {input: 5.00, output: 15.00}
hidden: true
gpt-5-mini:
provider: do-ai
upstream: openai-gpt-5-mini
fallbacks:
- provider: openai-direct
upstream: gpt-5-mini
pricing: {input: 1.25, output: 5.00}
hidden: true
gpt-5-nano:
provider: do-ai
upstream: openai-gpt-5-nano
fallbacks:
- provider: openai-direct
upstream: gpt-5-nano
pricing: {input: 0.30, output: 1.20}
hidden: true
gpt-5.1-codex-max:
provider: do-ai
upstream: openai-gpt-5.1-codex-max
fallbacks:
- provider: openai-direct
upstream: gpt-5.1-codex-max
pricing: {input: 3.00, output: 12.00}
hidden: true
gpt-5.2:
provider: do-ai
upstream: openai-gpt-5.2
fallbacks:
- provider: openai-direct
upstream: gpt-5.2
pricing: {input: 5.00, output: 15.00}
hidden: true
gpt-5.2-pro:
provider: do-ai
upstream: openai-gpt-5.2-pro
fallbacks:
- provider: openai-direct
upstream: gpt-5.2-pro
pricing: {input: 10.00, output: 30.00}
hidden: true
# ── OpenAI frontier via DO-AI (premium; positive balance required) ─────
# Enabled for customers 2026-07-18: verified live on inference.do-ai.run.
# Pricing mirrors the gpt-5 line (do-ai publishes no per-model rate).
gpt-5.4:
provider: do-ai
upstream: openai-gpt-5.4
context_window: 400000
max_output_tokens: 128000
vision: true # verified live
tools: true # verified live: emitted tool_calls
premium: true
pricing: {input: 5.00, output: 15.00}
gpt-5.5:
provider: do-ai
upstream: openai-gpt-5.5
context_window: 1000000
max_output_tokens: 128000
vision: true # verified live
tools: true # verified live
premium: true
pricing: {input: 5.00, output: 15.00}
# gpt-5.6 ships as three sibling variants (luna/sol/terra). All accept image
# input but REJECT function/tool calling on do-ai today (verified: HTTP 400 on
# tools across all three) — so `tools` is intentionally unset (honest: absent =
# not advertised). Revisit when do-ai enables tools for the 5.6 line.
gpt-5.6-luna:
provider: do-ai
upstream: openai-gpt-5.6-luna
context_window: 1050000
max_output_tokens: 128000
vision: true # verified live
premium: true
pricing: {input: 5.00, output: 15.00}
gpt-5.6-sol:
provider: do-ai
upstream: openai-gpt-5.6-sol
context_window: 1050000
max_output_tokens: 128000
vision: true
premium: true
pricing: {input: 5.00, output: 15.00}
gpt-5.6-terra:
provider: do-ai
upstream: openai-gpt-5.6-terra
context_window: 1050000
max_output_tokens: 128000
vision: true
premium: true
pricing: {input: 5.00, output: 15.00}
gpt-oss-120b:
context_window: 128000
provider: do-ai
upstream: openai-gpt-oss-120b
pricing: {input: 0.90, output: 0.90}
gpt-oss-20b:
context_window: 128000
provider: do-ai
upstream: openai-gpt-oss-20b
pricing: {input: 0.20, output: 0.20}
o1:
context_window: 200000
provider: do-ai
upstream: openai-o1
pricing: {input: 15.00, output: 60.00}
hidden: true
o3:
context_window: 200000
max_output_tokens: 100000
provider: do-ai
upstream: openai-o3
vision: true # verified live: answered image color
tools: true # verified live: emitted tool_calls
premium: true
pricing: {input: 10.00, output: 40.00}
o3-mini:
context_window: 200000
provider: do-ai
upstream: openai-o3-mini
pricing: {input: 1.10, output: 4.40}
hidden: true
# Anthropic via DO-AI
claude-3-5-haiku:
provider: do-ai
upstream: anthropic-claude-3.5-haiku
fallbacks:
- provider: anthropic
upstream: claude-3-5-haiku-20241022
pricing: {input: 0.80, output: 4.00}
claude-3-7-sonnet:
provider: do-ai
upstream: anthropic-claude-3.7-sonnet
fallbacks:
- provider: anthropic
upstream: claude-3-7-sonnet-20250219
pricing: {input: 3.00, output: 15.00}
claude-4-1-opus:
provider: do-ai
upstream: anthropic-claude-4.1-opus
fallbacks:
- provider: anthropic
upstream: claude-4-1-opus-20250620
pricing: {input: 15.00, output: 75.00}
hidden: true
claude-haiku-4-5:
provider: do-ai
upstream: anthropic-claude-haiku-4.5
context_window: 200000
max_output_tokens: 8192
vision: true # verified live: answered image color
tools: true # verified live: emitted tool_calls
fallbacks:
- provider: anthropic
upstream: claude-haiku-4-5-20251001
pricing: {input: 1.00, output: 5.00}
claude-opus-4:
provider: do-ai
upstream: anthropic-claude-opus-4
fallbacks:
- provider: anthropic
upstream: claude-opus-4-20250514
pricing: {input: 15.00, output: 75.00}
claude-opus-4-5:
provider: do-ai
upstream: anthropic-claude-opus-4.5
fallbacks:
- provider: anthropic
upstream: claude-opus-4-5-20250826
pricing: {input: 15.00, output: 75.00}
hidden: true
claude-opus-4-6:
provider: do-ai
upstream: anthropic-claude-opus-4.6
fallbacks:
- provider: anthropic
upstream: claude-opus-4-6-20250514
pricing: {input: 15.00, output: 75.00}
hidden: true
claude-opus-4-7:
provider: do-ai
upstream: anthropic-claude-opus-4.7
context_window: 200000
max_output_tokens: 8192
vision: true # verified live
tools: true # verified live
premium: true
fallbacks:
- provider: anthropic
upstream: claude-opus-4-7
pricing: {input: 15.00, output: 75.00}
claude-opus-4-8:
provider: do-ai
upstream: anthropic-claude-opus-4.8
context_window: 1000000
max_output_tokens: 128000
vision: true # verified live: answered image color
tools: true # verified live: emitted tool_calls
premium: true
fallbacks:
- provider: anthropic
upstream: claude-opus-4-8
pricing: {input: 15.00, output: 75.00}
claude-sonnet-4:
provider: do-ai
upstream: anthropic-claude-sonnet-4
fallbacks:
- provider: anthropic
upstream: claude-sonnet-4-20250514
pricing: {input: 3.00, output: 15.00}
claude-sonnet-4-5:
provider: do-ai
upstream: anthropic-claude-4.5-sonnet
fallbacks:
- provider: anthropic
upstream: claude-sonnet-4-5-20250929
pricing: {input: 3.00, output: 15.00}
hidden: true
# claude-sonnet-4-6 is the default model for Hanzo bot agents. Now routes to the
# REAL anthropic-claude-4.6-sonnet upstream — verified served on this account
# 2026-07-18 (2xx chat+vision+tools); it previously degraded to 4.5 while 4.6
# 403'd. anthropic-direct failover stays on the working 4.5 id (that provider is
# disabled by default; a 4.5 safety net beats a guessed 4.6 direct id).
claude-sonnet-4-6:
provider: do-ai
upstream: anthropic-claude-4.6-sonnet
context_window: 200000
max_output_tokens: 8192
vision: true # verified live: accepted image input
tools: true # verified live: emitted tool_calls
fallbacks:
- provider: anthropic
upstream: claude-sonnet-4-5-20250929
pricing: {input: 3.00, output: 15.00}
# ── Anthropic frontier via DO-AI (premium) — enabled 2026-07-18 ────────
# claude-sonnet-5: 1M-context Sonnet. Chat + tools verified live; it REJECTS
# image input on do-ai today (HTTP 400 on image_url — verified twice), so
# `vision` is intentionally unset (honest: absent = not advertised).
claude-sonnet-5:
provider: do-ai
upstream: anthropic-claude-5-sonnet
context_window: 1000000
max_output_tokens: 128000
tools: true # verified live: emitted tool_calls
premium: true
pricing: {input: 3.00, output: 15.00}
# claude-fable-5: Anthropic frontier creative model, 1M context. Chat + vision +
# tools all verified live. Pricing mirrors the opus frontier tier (do-ai
# publishes no per-model rate); revisit if a published rate appears.
claude-fable-5:
provider: do-ai
upstream: anthropic-claude-fable-5
context_window: 1000000
max_output_tokens: 128000
vision: true # verified live: answered image color
tools: true # verified live: emitted tool_calls
premium: true
pricing: {input: 15.00, output: 75.00}
# Open source via DO-AI
deepseek-r1-distill-70b:
context_window: 131072 # MEASURED (same upstream as deepseek-r1-distill-llama-70b)
provider: do-ai
upstream: deepseek-r1-distill-llama-70b
pricing: {input: 0.35, output: 1.20}
deepseek-v3.2:
context_window: 131072
provider: do-ai
upstream: deepseek-3.2
pricing: {input: 0.27, output: 1.10}
deepseek-reasoner:
provider: do-ai
upstream: deepseek-r1
pricing: {input: 3.00, output: 8.00}
gemma-3-31b:
provider: do-ai
upstream: gemma-3-31b
pricing: {input: 0.10, output: 0.30}
gemma-4-31b:
context_window: 256000
provider: do-ai
upstream: gemma-4-31B-it # real DO catalog id (was "gemma-4-31b" → 404)
pricing: {input: 0.18, output: 0.50}
kimi-k2:
provider: do-ai
upstream: kimi-k2
pricing: {input: 2.00, output: 8.00}
mistral-3-14b:
provider: do-ai
upstream: mistral-3-14b
pricing: {input: 0.08, output: 0.24}
mistral-small:
provider: do-ai
upstream: mistral-small
pricing: {input: 0.10, output: 0.30}
nemotron-3-nano:
provider: do-ai
upstream: nemotron-3-nano
pricing: {input: 0.04, output: 0.12}
nemotron-nano-vl:
context_window: 128000
provider: do-ai
upstream: nemotron-nano-12b-v2-vl
pricing: {input: 0.08, output: 0.24}
qwen3-coder:
provider: do-ai
upstream: qwen3-coder-480b
pricing: {input: 3.60, output: 3.60}
llama-3.1-8b:
provider: do-ai
upstream: llama3-8b-instruct
pricing: {input: 0.10, output: 0.10}
llama-3.3-70b:
context_window: 128000
provider: do-ai
upstream: llama3.3-70b-instruct
pricing: {input: 0.59, output: 0.79}
mistral-nemo:
provider: do-ai
upstream: mistral-nemo-instruct-2407
pricing: {input: 0.15, output: 0.15}
qwen3-32b:
context_window: 32768
provider: do-ai
upstream: alibaba-qwen3-32b
hidden: true
pricing: {input: 0.20, output: 0.60}
glm-5.2:
provider: do-ai
upstream: glm-5.2
premium: true
context_window: 262144 # measured
max_output_tokens: 52429
tools: true # verified live: emitted tool_calls (rejects image input → no vision)
pricing: {input: 3.00, output: 9.60}
# ── Strong open models via DO-AI — enabled for customers 2026-07-18 ────
# deepseek-v4-pro and kimi-k2.6 are exposed by unhiding their upstream-leaf
# entries below (the leaf key already IS the clean public name). llama-4-maverick
# and qwen3.5-397b get their own SKUs here. All verified live. Prices align to
# each family's existing sibling (do-ai publishes no per-model rate) — estimates.
llama-4-maverick:
provider: do-ai
upstream: llama-4-maverick
context_window: 128000
max_output_tokens: 16384
vision: true # verified live: answered image color (Llama 4 is multimodal)
tools: true # verified live: emitted tool_calls
pricing: {input: 0.35, output: 1.15} # estimate, aligned to the llama family tier
qwen3.5-397b:
provider: do-ai
upstream: qwen3.5-397b-a17b
context_window: 131072
max_output_tokens: 26214
tools: true # verified live: emitted tool_calls (vision unconfirmed → not advertised)
pricing: {input: 0.60, output: 2.20} # estimate, aligned to the qwen MoE tier
# ── DO-AI aliases (hidden, still callable) ─────────────────────────────
openai/gpt-4o:
context_window: 128000
provider: do-ai
upstream: openai-gpt-4o
hidden: true
alias_pricing: gpt-4o
openai/gpt-4o-mini:
context_window: 128000
provider: do-ai
upstream: openai-gpt-4o-mini
hidden: true
alias_pricing: gpt-4o-mini
openai/gpt-5:
context_window: 400000
provider: do-ai
upstream: openai-gpt-5
hidden: true
alias_pricing: gpt-5
openai/o3:
context_window: 200000
provider: do-ai
upstream: openai-o3
hidden: true
alias_pricing: o3
openai/o3-mini:
context_window: 200000
provider: do-ai
upstream: openai-o3-mini
hidden: true
alias_pricing: o3-mini
anthropic/claude-haiku-4-5-20251001:
context_window: 200000
provider: do-ai
upstream: anthropic-claude-haiku-4.5
hidden: true
alias_pricing: claude-haiku-4-5
anthropic/claude-opus-4-6:
context_window: 200000
provider: do-ai
upstream: anthropic-claude-opus-4.6
hidden: true
alias_pricing: claude-opus-4-6
anthropic/claude-sonnet-4-5-20250929:
context_window: 200000
provider: do-ai
upstream: anthropic-claude-4.5-sonnet
hidden: true
alias_pricing: claude-sonnet-4-5
# ── Fireworks premium models (hidden, still callable) ──────────────────
fireworks/cogito-671b:
provider: fireworks
upstream: accounts/cogito/models/cogito-671b-v2-p1
premium: true
hidden: true
pricing: {input: 3.00, output: 3.00}
fireworks/deepseek-v3p1:
provider: fireworks
upstream: accounts/fireworks/models/deepseek-v3p1
premium: true
hidden: true
pricing: {input: 0.90, output: 0.90}
fireworks/deepseek-v3p2:
provider: fireworks
upstream: accounts/fireworks/models/deepseek-v3p2
premium: true
hidden: true
pricing: {input: 0.90, output: 0.90}
fireworks/glm-4p7:
provider: fireworks
upstream: accounts/fireworks/models/glm-4p7
premium: true
hidden: true
pricing: {input: 1.80, output: 6.60}
fireworks/glm-5:
provider: fireworks
upstream: accounts/fireworks/models/glm-5
premium: true
hidden: true
pricing: {input: 3.00, output: 9.60}
fireworks/gpt-oss-120b:
provider: fireworks
upstream: accounts/fireworks/models/gpt-oss-120b
premium: true
hidden: true
pricing: {input: 0.90, output: 0.90}
fireworks/gpt-oss-20b:
provider: fireworks
upstream: accounts/fireworks/models/gpt-oss-20b
premium: true
hidden: true
pricing: {input: 0.20, output: 0.20}
fireworks/kimi-k2:
provider: fireworks
upstream: accounts/fireworks/models/kimi-k2-instruct-0905
premium: true
hidden: true
pricing: {input: 2.00, output: 8.00}
fireworks/kimi-k2-thinking:
provider: fireworks
upstream: accounts/fireworks/models/kimi-k2-thinking
premium: true
hidden: true
pricing: {input: 2.00, output: 8.00}
fireworks/kimi-k2p5:
provider: fireworks
upstream: accounts/fireworks/models/kimi-k2p5
premium: true
hidden: true
pricing: {input: 2.50, output: 10.00}
fireworks/llama-3.3-70b:
provider: fireworks
upstream: accounts/fireworks/models/llama-v3p3-70b-instruct
premium: true
hidden: true
pricing: {input: 0.59, output: 0.79}
fireworks/minimax-m2p1:
provider: fireworks
upstream: accounts/fireworks/models/minimax-m2p1
premium: true
hidden: true
pricing: {input: 1.00, output: 4.00}
fireworks/minimax-m2p5:
provider: fireworks
upstream: accounts/fireworks/models/minimax-m2p5
premium: true
hidden: true
pricing: {input: 1.00, output: 4.00}
fireworks/mixtral-8x22b:
provider: fireworks
upstream: accounts/fireworks/models/mixtral-8x22b-instruct
premium: true
hidden: true
pricing: {input: 0.90, output: 0.90}
fireworks/qwen3-8b:
provider: fireworks
upstream: accounts/fireworks/models/qwen3-8b
premium: true
hidden: true
pricing: {input: 0.60, output: 0.60}
fireworks/qwen3-vl-30b:
provider: fireworks
upstream: accounts/fireworks/models/qwen3-vl-30b-a3b-instruct
premium: true
hidden: true
pricing: {input: 0.45, output: 1.80}
fireworks/qwen3-vl-30b-thinking:
provider: fireworks
upstream: accounts/fireworks/models/qwen3-vl-30b-a3b-thinking
premium: true
hidden: true
pricing: {input: 0.45, output: 1.80}
# Pricing-only Fireworks entries (no route, just for cost calculation)
fireworks/deepseek-r1:
provider: fireworks
upstream: accounts/fireworks/models/deepseek-r1
premium: true
hidden: true
pricing_only: true
pricing: {input: 3.00, output: 8.00}
fireworks/deepseek-v3:
provider: fireworks
upstream: accounts/fireworks/models/deepseek-v3
premium: true
hidden: true
pricing_only: true
pricing: {input: 0.90, output: 0.90}
fireworks/glm-5-thinking:
provider: fireworks
upstream: accounts/fireworks/models/glm-5-thinking
premium: true
hidden: true
pricing_only: true
pricing: {input: 3.00, output: 9.60}
fireworks/qwen3-4b:
provider: fireworks
upstream: accounts/fireworks/models/qwen3-4b
premium: true
hidden: true
pricing_only: true
pricing: {input: 0.30, output: 0.30}
fireworks/qwen3-235b-a22b:
provider: fireworks
upstream: accounts/fireworks/models/qwen3-235b-a22b
premium: true
hidden: true
pricing_only: true
pricing: {input: 3.60, output: 3.60}
fireworks/qwen3-coder-30b-a3b:
provider: fireworks
upstream: accounts/fireworks/models/qwen3-coder-30b-a3b
premium: true
hidden: true
pricing_only: true
pricing: {input: 1.50, output: 1.50}
fireworks/qwen3-coder-480b:
provider: fireworks
upstream: accounts/fireworks/models/qwen3-coder-480b
premium: true
hidden: true
pricing_only: true
pricing: {input: 3.60, output: 3.60}
fireworks/qwen3-coder-480b-bf16:
provider: fireworks
upstream: accounts/fireworks/models/qwen3-coder-480b-bf16
premium: true
hidden: true
pricing_only: true
pricing: {input: 4.50, output: 4.50}
fireworks/qwen3-next-80b-a3b:
provider: fireworks
upstream: accounts/fireworks/models/qwen3-next-80b-a3b
premium: true
hidden: true
pricing_only: true
pricing: {input: 2.70, output: 2.70}
fireworks/qwen3-next-80b-a3b-thinking:
provider: fireworks
upstream: accounts/fireworks/models/qwen3-next-80b-a3b-thinking
premium: true
hidden: true
pricing_only: true
pricing: {input: 2.70, output: 2.70}
fireworks/qwen3-vl-30b-a3b:
provider: fireworks
upstream: accounts/fireworks/models/qwen3-vl-30b-a3b
premium: true
hidden: true
pricing_only: true
pricing: {input: 0.45, output: 1.80}
fireworks/qwen3-vl-235b:
provider: fireworks
upstream: accounts/fireworks/models/qwen3-vl-235b
premium: true
hidden: true
pricing_only: true
pricing: {input: 1.20, output: 1.20}
# ── OpenAI Direct premium models (hidden, use top-level names) ─────────
openai-direct/gpt-4o:
provider: openai-direct
upstream: gpt-4o
premium: true
hidden: true
pricing: {input: 2.50, output: 10.00}
openai-direct/gpt-4o-mini:
provider: openai-direct
upstream: gpt-4o-mini
premium: true
hidden: true
pricing: {input: 0.15, output: 0.60}
openai-direct/gpt-5:
provider: openai-direct
upstream: gpt-5
premium: true
hidden: true
pricing: {input: 5.00, output: 15.00}
openai-direct/o3:
provider: openai-direct
upstream: o3
premium: true
hidden: true
pricing: {input: 10.00, output: 40.00}
openai-direct/o3-mini:
provider: openai-direct
upstream: o3-mini
premium: true
hidden: true
pricing: {input: 1.10, output: 4.40}
# ── Anthropic Direct premium models (hidden, use top-level names) ─────
anthropic/claude-haiku-4-5:
provider: anthropic
upstream: claude-haiku-4-5-20251001
premium: true
hidden: true
pricing: {input: 1.00, output: 5.00}
anthropic/claude-sonnet-4:
provider: anthropic
upstream: claude-sonnet-4-20250514
premium: true
hidden: true
pricing: {input: 3.00, output: 15.00}
anthropic/claude-opus-4:
provider: anthropic
upstream: claude-opus-4-20250514
premium: true
hidden: true
pricing: {input: 15.00, output: 75.00}
# do-ai fal-hosted diffusion via the async image API (model/doai_image.go).
# FLUX.1 schnell is the flagship/fast model; SDXL serves the -sdxl/-ssd
# variants. Pricing is per-token-shaped here only because the config schema
# is token-based; image calls bill per image via controllers.imageCostCents.
# ── DO-AI image: Stable Diffusion 3.5 Large ───────────────────────────
# Unlike the fal FLUX/SDXL models (async-invoke), SD 3.5 Large is served on
# the SYNCHRONOUS OpenAI /images/generations shape. model.GenerateImageForModel
# dispatches it via client.Images.Generate; getOpenAiModelType classifies it
# as an image model. Usable via /v1/images/generations and the chat-image path.
# Billed $0.08/image (controllers.imageCostCents) — token pricing is unused.
stable-diffusion-3.5-large:
context_window: 256
provider: do-ai
upstream: stable-diffusion-3.5-large
# ── DO-AI text-to-video (served at POST /v1/videos/generations) ────────
# wan2-2-t2v-a14b is the only text-to-video model in the do-ai catalog,
# served on the OpenAI Sora-style async /v1/videos API (create → poll →
# download) by model/doai_video.go. getOpenAiModelType classifies it via
# isDOAIVideoModel → the videos path. Billed per video via
# controllers.videoCostCents (40¢/video, DO publishes no per-video rate) —
# token pricing is unused for video. The unbranded raw id below stays on do-ai;
# provider (same async /v1/videos API), owned_by hanzo (upstream never exposed).
#
# premium: true on EVERY video id (including this unbranded passthrough) tags the
# usage record as premium; a t2v inference is minutes of A14B GPU compute (~40¢),
# billed per video via controllers.videoCostCents through the one metering path.
# Access is gated purely on a positive prepaid balance (there is no free credit),
# so a $0 signup is refused at the balance gate before ever reaching video. This
# YAML is the runtime source of truth (it wins over the static route map), so the
# provider re-point and the flag have to live here too, not only in
# controllers/model_routes.go.
wan2-2-t2v-a14b:
context_window: 100
provider: do-ai
upstream: wan2-2-t2v-a14b
premium: true
# ── DO-AI Inference Router (served at POST /v1/chat/completions) ───────
# DO's model router auto-selects the best foundation model per request and
# returns the chosen model in the response. Callable as plain chat. The router
# itself adds no cost (public preview); the request bills at the underlying
# model it selects — the price below is an in-pattern estimate for that class
# (matching gpt-oss-120b), NOT a published per-router rate.
"router:general":
provider: do-ai
upstream: "router:general"
pricing: {input: 0.90, output: 0.90}
"router:knowledge-base-document":
provider: do-ai
upstream: "router:knowledge-base-document"
pricing: {input: 0.90, output: 0.90}
"router:software-engineering":
provider: do-ai
upstream: "router:software-engineering"
pricing: {input: 0.90, output: 0.90}
"router:software-engineering-01":
provider: do-ai
upstream: "router:software-engineering-01"
pricing: {input: 0.90, output: 0.90}
"router:writing":
provider: do-ai
upstream: "router:writing"
pricing: {input: 0.90, output: 0.90}
#
# hidden: it does not appear in /v1/models, so no user can discover or select
# it, but it stays callable for anyone we deliberately let through (waitlist /
# allowlist). DO currently 403s it for this account; the moment the agreement
# lands this serves 1M with no further change, and dropping `hidden` is what
# makes it public. Until then it degrades to the same 1M we can actually serve
# rather than failing, so an allowlisted caller is never broken.
# the upstream (Kimi K2.7 Code when it lands; qwen3-coder today) in yaml only.
# (qwen3.5-397b-a17b, 397B, 131K context) when K2.6 is unavailable.
# ── OpenAI-compatible embeddings (served at POST /v1/embeddings) ───────
# Canonical text-embedding-* names route to OpenAI Direct, whose key
# (kms://OPENAI_API_KEY) is already configured for chat — so the OpenAI SDK
# works unchanged with NO new secret. Non-premium; CPU-only
# (provider-routed, no GPU node).
text-embedding-3-small:
provider: openai-direct
upstream: text-embedding-3-small
owned_by: openai
pricing: {input: 0.02, output: 0.00}
text-embedding-3-large:
provider: openai-direct
upstream: text-embedding-3-large
owned_by: openai
pricing: {input: 0.13, output: 0.00}
text-embedding-ada-002:
provider: openai-direct
upstream: text-embedding-ada-002
owned_by: openai
pricing: {input: 0.10, output: 0.00}
# ── DO-AI embeddings (served at POST /v1/embeddings) ──────────────────
# Open sentence-transformer / BGE / GTE / E5 embedding models served by the
# do-ai provider (the same working key as chat). Non-premium.
# Prices are DigitalOcean's published per-1M embedding rates;
# embeddings bill on input tokens only (output 0). owned_by = do-ai provider.
bge-m3:
provider: do-ai
upstream: bge-m3
pricing: {input: 0.02, output: 0.00}
e5-large-v2:
provider: do-ai
upstream: e5-large-v2
pricing: {input: 0.02, output: 0.00}
gte-large-en-v1.5:
provider: do-ai
upstream: gte-large-en-v1.5
pricing: {input: 0.09, output: 0.00}
all-mini-lm-l6-v2:
provider: do-ai
upstream: all-mini-lm-l6-v2
pricing: {input: 0.009, output: 0.00}
multi-qa-mpnet-base-dot-v1:
provider: do-ai
upstream: multi-qa-mpnet-base-dot-v1
pricing: {input: 0.009, output: 0.00}
# ── Rerank (served at POST /v1/rerank) ────────────────────────────────
# this embedding model (text-embedding-3-large on OpenAI Direct) and ranks by
# cosine similarity — a real ranked response with no rerank-specific key and no
# GPU. To add a cross-encoder reranker, seed a Jina/Cohere provider (drop
# JINA_API_KEY / COHERE_API_KEY into the cloud-api-llm-keys KMS secret) and add
# a model here routed to it; /v1/rerank proxies native rerank providers directly.
# direct-Fireworks routes.
# Vision-language size variants
# virtual SKUs with no HF weights and no working upstream — removed
# Speech recognition (ASR)
# Text-to-speech (TTS)
# no working upstream — removed 2026-05-30.
# ── DO upstream leaves ──────────────────────────────────────────────
# Declared so every alias can inherit its upstream's REAL served window
# (ModelConfig.ContextWindow follows the alias chain). Windows come from
# the live DO catalog, not a guess. hidden: they are the plumbing behind
alibaba-qwen3-32b:
provider: do-ai
upstream: alibaba-qwen3-32b
context_window: 40960 # MEASURED. DO advertises 32,768.
hidden: true
anthropic-claude-4.1-opus:
provider: do-ai
upstream: anthropic-claude-4.1-opus
context_window: 200000
hidden: true
anthropic-claude-4.5-sonnet:
provider: do-ai
upstream: anthropic-claude-4.5-sonnet
context_window: 200000
hidden: true
anthropic-claude-haiku-4.5:
provider: do-ai
upstream: anthropic-claude-haiku-4.5
context_window: 200000
hidden: true
anthropic-claude-opus-4.5:
provider: do-ai
upstream: anthropic-claude-opus-4.5
context_window: 200000
hidden: true
anthropic-claude-opus-4.6:
provider: do-ai
upstream: anthropic-claude-opus-4.6
context_window: 200000
hidden: true
anthropic-claude-opus-4.7:
provider: do-ai
upstream: anthropic-claude-opus-4.7
context_window: 200000
hidden: true
anthropic-claude-opus-4.8:
provider: do-ai
upstream: anthropic-claude-opus-4.8
context_window: 1000000
hidden: true
deepseek-3.2:
provider: do-ai
upstream: deepseek-3.2
context_window: 163840 # MEASURED. DO advertises 131,072.
hidden: true
deepseek-4-flash:
provider: do-ai
upstream: deepseek-4-flash
context_window: 262144 # MEASURED. DO advertises 65,536.
hidden: true
deepseek-r1-distill-llama-70b:
provider: do-ai
upstream: deepseek-r1-distill-llama-70b
context_window: 131072 # MEASURED. DO advertises 32,768.
hidden: true
# deepseek-v4-pro: VISIBLE customer SKU (enabled 2026-07-18). Chat + tools
# verified live; REJECTS image input on do-ai (HTTP 400) so `vision` stays unset.
deepseek-v4-pro:
provider: do-ai
upstream: deepseek-v4-pro
context_window: 1000000 # MEASURED: accepted 1,000,005 tokens. DO advertises 87,040 and caps
# every other model at 262,144 — this one is not capped.
# (>1.2M fails as "invalid request body" = HTTP payload limit,
# not a context limit — the model itself never refuses.)
max_output_tokens: 52429
tools: true # verified live: emitted tool_calls
pricing: {input: 0.90, output: 2.20} # estimate, aligned to the deepseek flagship tier
gemma-4-31B-it:
provider: do-ai
upstream: gemma-4-31B-it
context_window: 256000
hidden: true
glm-5:
provider: do-ai
upstream: glm-5
context_window: 64000
hidden: true
# kimi-k2.6: VISIBLE customer SKU (enabled 2026-07-18). Chat + tools verified
# live. Window kept at the measured 96000 (conservative — do-ai advertises
# 262144; under-claiming never breaks a request, over-claiming does).
kimi-k2.6:
provider: do-ai
upstream: kimi-k2.6
context_window: 96000
max_output_tokens: 52429
tools: true # verified live: emitted tool_calls
pricing: {input: 2.00, output: 8.00} # aligned to the kimi-k2 sibling rate
llama3.3-70b-instruct:
provider: do-ai
upstream: llama3.3-70b-instruct
context_window: 128000
hidden: true
minimax-m2.5:
provider: do-ai
upstream: minimax-m2.5
context_window: 65536
hidden: true
nemotron-nano-12b-v2-vl:
provider: do-ai
upstream: nemotron-nano-12b-v2-vl
context_window: 128000
hidden: true
openai-gpt-4.1:
provider: do-ai
upstream: openai-gpt-4.1
context_window: 1047576
hidden: true
openai-gpt-4o:
provider: do-ai
upstream: openai-gpt-4o
context_window: 128000
hidden: true
openai-gpt-4o-mini:
provider: do-ai
upstream: openai-gpt-4o-mini
context_window: 128000
hidden: true
openai-gpt-5:
provider: do-ai
upstream: openai-gpt-5
context_window: 400000
hidden: true
openai-gpt-5-mini:
provider: do-ai
upstream: openai-gpt-5-mini
context_window: 400000
hidden: true
openai-gpt-5-nano:
provider: do-ai
upstream: openai-gpt-5-nano
context_window: 400000
hidden: true
openai-gpt-5.1-codex-max:
provider: do-ai
upstream: openai-gpt-5.1-codex-max
context_window: 400000
hidden: true
openai-gpt-5.2:
provider: do-ai
upstream: openai-gpt-5.2
context_window: 400000
hidden: true
openai-gpt-5.2-pro:
provider: do-ai
upstream: openai-gpt-5.2-pro
context_window: 400000
hidden: true
openai-gpt-oss-120b:
provider: do-ai
upstream: openai-gpt-oss-120b
context_window: 128000
hidden: true
openai-gpt-oss-20b:
provider: do-ai
upstream: openai-gpt-oss-20b
context_window: 128000
hidden: true
openai-o1:
provider: do-ai
upstream: openai-o1
context_window: 200000
hidden: true
openai-o3:
provider: do-ai
upstream: openai-o3
context_window: 200000
hidden: true
openai-o3-mini:
provider: do-ai
upstream: openai-o3-mini
context_window: 200000
hidden: true
qwen3-coder-flash:
provider: do-ai
upstream: qwen3-coder-flash
context_window: 262144 # measured
hidden: true
qwen3.5-397b-a17b:
provider: do-ai
upstream: qwen3.5-397b-a17b
context_window: 131072
hidden: true
# Anthropic frontier leaf. Advertised (and, once entitled, served) at 1M.
# 403 for this account today — present so the aliases above resolve.
anthropic-claude-fable-5:
provider: do-ai
upstream: anthropic-claude-fable-5
context_window: 1000000
hidden: true
nemotron-3-nano-omni:
provider: do-ai
upstream: nemotron-3-nano-omni
hidden: true
context_window: 65536