Models

For someone running uptide fix with an API key who wants to choose the provider, the model and the budget, or point at their own endpoint. Measured results per provider are in the provider evaluation; rates in model pricing.

Choosing a provider and model

Assisted fixes support Anthropic, OpenAI and Gemini (experimental), with the same prompts, tool, verification and publish gate. Set the chosen provider’s key in your environment: ANTHROPIC_API_KEY, OPENAI_API_KEY or GEMINI_API_KEY. Never put keys in repository config or command arguments.

npx uptide fix zod --provider openai --model gpt-6.1-sol
UPTIDE_PROVIDER=gemini UPTIDE_MODEL=gemini-3.8-flash uptide fix stripe
npx uptide fix zod --max-cost 1
npx uptide fix zod --no-llm

For each setting, precedence is flag → UPTIDE_PROVIDER / UPTIDE_MODEL → nearest uptide.config.json up to the Git root. Without a provider setting, detection checks ANTHROPIC_API_KEY, then OPENAI_API_KEY, then GEMINI_API_KEY; Anthropic remains the overall default. An explicitly selected provider never falls back to another provider. The selected provider/model prints before the fix starts, and spend prints at the end.

The optional config accepts only provider and model strings:

{ "provider": "openai", "model": "gpt-6.1-sol" }

Unknown fields, nested settings and key-like values are rejected, including when flags would override them. Keys are read only from the environment. No key: a generic fix exits before cloning, installing or creating a branch; migration packs still apply rule-based fixes and leave assisted sites manual. --no-llm disables all model calls.

Defaults

Checked against official documentation on 2026-10-04:

Provider	Default	Reference
Anthropic	claude-sonnet-5-5 (medium effort)	Sonnet 5.5; selected after the storefront comparison
OpenAI	gpt-6.1-sol	GPT-6.1 Sol, through the Responses API
Gemini	gemini-3.8-flash	Gemini 3.8 Flash

Measured on the public storefront fixture (2026-10-04; USD from returned usage). Each cell lists zod / Stripe; every run has a $1 ceiling:

Provider	Model	Verified	Agent sites	Attempts	Cost USD
Anthropic	claude-sonnet-5-5 (medium)	pass / pass	4 / 3	2 / 2	$0.013244 / $0.021592
OpenAI	gpt-6.1-sol	pass / pass	4 / 3	2 / 3	$0.006806 / $0.018677
Gemini	gemini-3.8-flash	fail / fail	0 / 0	12 / 9	$0.000000 / $0.000000

Gemini’s serial reruns failed after HTTP 503/429 and transport errors, despite a successful protocol smoke. Its zero confirmed cost excludes $0.473901 / $0.368667 retained for unreported usage; it is not a claim of zero possible billing. Full trials, reservations, times, exact echoed IDs and the Sonnet 4.6/high/medium comparison are in the provider evaluation. These single fixture trials are not a general model ranking.

Tool choice follows each model’s capabilities. Sonnet 5.5 uses auto with a strict submit_patch schema and a tool-only system instruction; Sonnet 4.6, OpenAI and Gemini use forced tool calls. Missing or invalid calls consume an attempt and their reported cost, then receive the feedback “respond only by calling submit_patch”. HTTP 429 is reported as rate limiting, waits for Retry-After (60 seconds when absent), and retries only when another reservation fits the budget.

--max-cost

--max-cost defaults to $1 per package, including zod and stripe. Before every call and retry, Uptide reserves its worst-case input and maximum output cost. A call that would exceed the remaining budget is never sent. Actual returned usage replaces the reservation; failures without valid usage keep the full reservation, shown separately as unknown spend. Rejected patches still consume budget. Unfinished sites remain manual and cannot pass the publish gate. Unknown models use the provider’s highest listed rates with a warning. See model pricing for rates, token estimation and the scope of accounting.

OPENAI_BASE_URL

OPENAI_BASE_URL selects an OpenAI Chat Completions-compatible, not verified endpoint. Uptide appends /chat/completions to the supplied API base URL and uses Bearer authentication, strict functions and forced submit_patch. The official OpenAI endpoint uses Responses when this override is absent. HTTPS is required, except HTTP localhost for local servers. Custom deployments must support this protocol and the supplied model ID; Azure-specific authentication and routing are not inferred. Custom endpoint prices may differ from the built-in OpenAI table.

Privacy

Assisted fixes send the finding, enclosing code snippet and compiler error to the provider you chose (or your OPENAI_BASE_URL). Your provider’s data policy applies. --no-llm keeps assisted fixes off. Anonymous telemetry, when enabled, adds only the provider and a public model ID; private/custom model IDs become custom. The full model: privacy.