Models & Providers

Tracker runs any model through one of three backends, prices every run against a shared catalog, and lets you route traffic straight to a provider or through a gateway. The model list and prices below are derived from dippin-lang (the pinned version) — a single source of truth, so what you see here is exactly what the engine uses to estimate cost and enforce --max-cost.

Backends

Every agent node runs on one backend; you can mix them per node.

BackendWhat it isAuth
native (default)Tracker’s own turn loop — Anthropic, OpenAI, Google Gemini, and any OpenAI-compatible endpoint. Tool calls, streaming, context compaction, and the writable_paths jail all live here.Provider API keys (below)
claude-codeSpawns the claude CLI as a subprocess. API keys are stripped so it uses your Max/Pro subscription auth.Subscription (OAuth)
acpACP-protocol client for headless external agents (claude-agent-acp / codex-acp / gemini --acp).Per-agent

Full backend internals: Architecture.

Native providers & keys

The native backend speaks four provider APIs directly. Set the matching environment variable (or configure it with tracker setup); tracker doctor verifies them.

ProviderKey env varBase-URL override
AnthropicANTHROPIC_API_KEYANTHROPIC_BASE_URL
OpenAIOPENAI_API_KEYOPENAI_BASE_URL
Google (Gemini)GEMINI_API_KEYGEMINI_BASE_URL
OpenAI-compatibleOPENAI_API_KEY (or provider’s)<PROVIDER>_BASE_URL

OpenAI-compatible is how you reach the other providers in the catalog below (DeepSeek, Mistral, xAI/Grok, Z.AI, MiniMax, Moonshot, …): point a *_BASE_URL at their OpenAI-compatible endpoint. A per-provider <PROVIDER>_BASE_URL always wins over the gateway.

Gateway routing

Instead of talking to each provider directly, you can route everything through a single gateway — a Cloudflare AI Gateway or the 2389 Bedrock proxy — by setting one URL. Tracker appends a per-provider suffix and otherwise passes the provider’s API/SSE through unchanged, so a gateway is transparent.

SettingMeaning
TRACKER_GATEWAY_URLBase URL of the gateway. When set, providers route through it.
TRACKER_GATEWAY_KINDcf-aig (default) or bedrock. Controls the per-provider suffix and routing rules.

cf-aig appends /anthropic, /openai, /google-ai-studio, or /compat per provider. bedrock routes Claude models to Amazon Bedrock (OpenAI/Gemini get a /v1 suffix; openai-compat is refused). Precedence: a provider *_BASE_URL beats the gateway. Setup + caveats: Architecture.

Supported models & prices

Every model dippin-lang prices, grouped by provider. Prices are US dollars per million tokens. Cached read is the discounted rate for a prompt-cache hit. Retired (first-party) means the model 404s on its first-party API but is still billable through a Bedrock/Vertex passthrough.

Tracker natively adapts Anthropic, OpenAI, and Gemini; the rest are reachable via an OpenAI-compatible base URL or a gateway. This table is generated from the pinned dippin version — it can never drift from what the engine bills.

Anthropic

ModelInput /MOutput /MCached read /MNotes
claude-fable-5$10.00$50.00$1.00
claude-fable-5-1$10.00$50.00$0.250
claude-haiku-3-5$0.800$4.00$0.080retired (first-party)
claude-haiku-4-5$1.00$5.00$0.100
claude-mythos-5$10.00$50.00$1.00
claude-opus-4-0$15.00$75.00$1.50retired (first-party)
claude-opus-4-1$15.00$75.00$1.50retired (first-party)
claude-opus-4-5$5.00$25.00$0.500
claude-opus-4-6$5.00$25.00$0.500
claude-opus-4-7$5.00$25.00$0.500
claude-opus-4-8$5.00$25.00$0.500
claude-opus-5$5.00$25.00$0.500
claude-sonnet-4-0$3.00$15.00$0.300retired (first-party)
claude-sonnet-4-5$3.00$15.00$0.300
claude-sonnet-4-6$3.00$15.00$0.300
claude-sonnet-5$3.00$15.00$0.300

OpenAI

ModelInput /MOutput /MCached read /MNotes
gpt-4.1$2.00$8.00$0.500
gpt-4.1-mini$0.400$1.60$0.100
gpt-4.1-nano$0.100$0.400$0.025
gpt-4o$2.50$10.00$1.25
gpt-4o-mini$0.150$0.600$0.075
gpt-5$1.25$10.00$0.125
gpt-5-mini$0.250$2.00$0.025
gpt-5-nano$0.050$0.400$0.005
gpt-5-pro$15.00$120.00$1.50
gpt-5.1$1.25$10.00$0.125
gpt-5.2$1.75$14.00$0.175
gpt-5.2-codex$1.75$14.00$0.175
gpt-5.2-pro$21.00$168.00$2.10
gpt-5.3-codex$1.75$14.00$0.175
gpt-5.4$2.50$15.00$0.250
gpt-5.4-mini$0.750$4.50$0.075
gpt-5.4-nano$0.200$1.25$0.020
gpt-5.4-pro$30.00$180.00$3.00
gpt-5.5$5.00$30.00$0.500
gpt-5.5-pro$30.00$180.00$3.00
gpt-5.6-luna$0.200$1.20$0.020
gpt-5.6-sol$4.00$20.00$0.400
gpt-5.6-terra$2.00$12.00$0.200
gpt-6-astra$10.00$50.00$1.00
o3$2.00$8.00$0.200
o3-mini$1.10$4.40$0.110
o3-pro$20.00$80.00$2.00
o4-mini$1.10$4.40$0.110

Google (Gemini)

ModelInput /MOutput /MCached read /MNotes
gemini-2.0-flash$0.100$0.400$0.010retired (first-party)
gemini-2.5-flash$0.300$2.50$0.030
gemini-2.5-flash-lite$0.100$0.400$0.010
gemini-2.5-pro$1.25$10.00$0.125
gemini-3-flash-preview$0.500$3.00$0.050
gemini-3.1-flash-lite$0.250$1.50$0.025
gemini-3.1-flash-lite-preview$0.250$1.50$0.025retired (first-party)
gemini-3.1-pro-preview$2.00$12.00$0.200
gemini-3.1-pro-preview-customtools$2.00$12.00$0.200
gemini-3.5-flash$1.50$9.00$0.150
gemini-3.5-flash-lite$0.300$2.50$0.030
gemini-3.6-flash$1.50$7.50$0.150
gemini-3.7-flash$1.50$7.50$0.150
gemini-3.8-flash$1.50$7.50$0.150

DeepSeek

ModelInput /MOutput /MCached read /MNotes
deepseek-chat$0.140$0.280
deepseek-reasoner$0.140$0.280
deepseek-v4-flash$0.140$0.280$0.003
deepseek-v4-pro$0.435$0.870$0.004
deepseek-v4-pro-0813$0.435$0.870$0.004

xAI (Grok)

ModelInput /MOutput /MCached read /MNotes
grok-4-1-fast-non-reasoning$1.25$2.50
grok-4-1-fast-reasoning$1.25$2.50
grok-4.20-0309-non-reasoning$1.25$2.50$0.200
grok-4.20-0309-reasoning$1.25$2.50$0.200
grok-4.20-multi-agent-0309$1.25$2.50$0.200
grok-4.3$1.25$2.50$0.200
grok-4.5$2.00$6.00$0.300
grok-4.6$2.00$6.00$0.500
grok-build-0.1$1.00$2.00$0.200

Mistral

ModelInput /MOutput /MCached read /MNotes
codestral$0.300$0.900$0.300
magistral-medium$2.00$5.00$2.00
ministral-3-14b-2512$0.200$0.200$0.200
ministral-3-3b-2512$0.100$0.100$0.100
ministral-3-8b-2512$0.150$0.150$0.150
ministral-8b$0.100$0.100$0.100retired (first-party)
mistral-large-3$0.500$1.50$0.500
mistral-medium-3$0.400$2.00$0.400
mistral-medium-3-1-2508$0.400$2.00$0.400retired (first-party)
mistral-medium-3-5-2604$1.50$7.50$1.50
mistral-nemo$0.020$0.040$0.020retired (first-party)
mistral-small$0.100$0.300$0.100
mistral-small-2603$0.100$0.300$0.100

Cohere

ModelInput /MOutput /MCached read /MNotes
command-a-03-2025$2.50$10.00$2.50
command-r$0.500$1.50$0.500retired (first-party)
command-r-08-2024$0.500$1.50$0.500
command-r-plus$2.50$10.00$2.50retired (first-party)
command-r-plus-08-2024$2.50$10.00$2.50
command-r7b$0.037$0.150$0.037
command-r7b-12-2024$0.037$0.150$0.037

Z.AI (GLM)

ModelInput /MOutput /MCached read /MNotes
glm-4.5$0.600$2.20$0.110
glm-4.5-air$0.200$1.10$0.030
glm-4.5-airx$1.10$4.50$0.220
glm-4.5-x$2.20$8.90$0.450
glm-4.6$0.600$2.20$0.110
glm-4.7$0.600$2.20$0.110
glm-4.7-flashx$0.070$0.400$0.010
glm-5$1.00$3.20$0.200
glm-5-turbo$1.20$4.00$0.240
glm-5.1$1.40$4.40$0.260
glm-5.2$1.40$4.40$0.260
glm-5.3$1.40$4.40$0.260
glm-5.3-flash$0.150$0.500$0.030

MiniMax

ModelInput /MOutput /MCached read /MNotes
MiniMax-M2$0.300$1.20
MiniMax-M2.1$0.300$1.20
MiniMax-M2.5$0.300$1.20
MiniMax-M2.5-highspeed$0.600$2.40
MiniMax-M2.7$0.300$1.20
MiniMax-M2.7-highspeed$0.600$2.40
MiniMax-M3$0.300$1.20

Moonshot (Kimi)

ModelInput /MOutput /MCached read /MNotes
kimi-k3$3.00$15.00$0.300

Meta (Muse)

ModelInput /MOutput /MCached read /MNotes
muse-spark-1.1$1.25$4.25$0.150
muse-spark-1.2$1.25$4.25$0.150
muse-spark-1.2-contributor$0.100$0.200$0.002
muse-spark-1.3$1.25$4.25$0.150
muse-spark-1.3-contributor$0.100$0.200$0.002

Cost governance

Because every model above carries a price, Tracker estimates the dollar cost of a run and enforces hard ceilings between nodes. Breach any cap and the run halts with a cost snapshot you can resume.

tracker --max-cost 500 --max-tokens 100000 --max-wall-time 30m build_product

--max-cost is in cents. A model dippin doesn’t price runs at $0 (with a one-time warning), so a ceiling can’t silently fail to apply. Full budget model: CLI Reference.