Register Gemini 3.6 Flash and 3.5 Flash-Lite (July 2026)

[blisspixel]

Adds the two GA-July-21-2026 Gemini models to the registry so primr can use them. Assessed against the model-selection policy (register, don't auto-default; defaults chosen on output $/1M).

Registered (available, not defaulted)

Model	in/out per 1M	Role fit
gemini-3.6-flash	$1.50 / $7.50	Successor to 3.5 Flash — same input, cheaper output + ~17% fewer output tokens, stronger coding/agent scores. Now the Pro-tier replacement candidate.
gemini-3.5-flash-lite	$0.30 / $2.50	Successor to 3.1 Flash-Lite; beats 3 Flash on benchmarks. Writing/utility candidate.

Deliberately not registered

• gemini-3.5-flash-cyber — limited-access dual-use pilot (governments/trusted partners via CodeMender), no public API/pricing, not primr's use case.
• gemini-3.5-pro — still partner-testing, not GA. The header note tracks it; register when the API slug goes live.

Policy + safety

• No default tier changed — a FLASH/PRO repoint stays eval-gated. test_model_registry_gemini.py pins both halves: the new ids are priced (so the mandatory cost gate can't KeyError) and are not the active FLASH/PRO defaults.
• Prices are the announced public rates; cached-input rates weren't published at launch, so they're left unset (conservative for the estimate gate) rather than guessed.
• $0 to add — registering a model is not running it.

Verified

• New ids resolve via get_model_config and price via calculate_cost without error.
• tests/test_config/ + test_cost_estimator + test_v123_provider_foundation + test_architecture — 266 passed; new gemini test 4 passed; ruff clean.