04.1 — admin / llm models
LLM Models registry.
{{ active_count }} active · {{ archived_count }} archived · {{ rows | length }} shown under the current filters
{% if just_created %}[ok]// model
{% endif %}
{% if just_updated %}{{ just_created }} created[ok]// model
{% endif %}
{% if just_archived %}{{ just_updated }} updated[ok]// model
{% endif %}
{% if just_restored %}{{ just_archived }} archived[ok]// model
{% endif %}
{% if assigned %}{{ just_restored }} restored[ok]// model assignment updated for
{% endif %}
{# ── FILTERS ──────────────────────────────────────────────── #}
{%- macro flink(label, params, active) -%}
{{ label }}
{%- endmacro -%}
{%- set ff = filter_frontier_only -%}
{%- set fr = filter_reasoning_only -%}
{# ── TABLE ────────────────────────────────────────────────── #}
{%- macro price(value) -%}
{%- if value is none or value == '' -%}
—
{%- else -%}
${{ '%0.2f' % (value | float) }}
{%- endif -%}
{%- endmacro -%}
{{ assigned }}
registry
{{ rows | length }} rows
{% if not rows %}
{% endif %}
// no models match the current filters.
{% else %}
| company {{ tip("Vendor / lab that ships the model. Display-only grouping.") }} | model id {{ tip("SDK identifier (e.g. claude-opus-4-7). Unique; the join key the resolver and llm_calls use.") }} | ver {{ tip("Human version label (e.g. 4.7 or 2025-01). Free text, optional.") }} | flags {{ tip("F = frontier (top-tier capability). R = reasoning (extended thinking). — = neither.") }} | in $/m {{ tip("Input price in USD per 1,000,000 input tokens.") }} | out $/m {{ tip("Output price in USD per 1,000,000 output tokens.") }} | cache r {{ tip("Cache-read price per 1M tokens. For Anthropic this is cache_read_price_per_million; OpenAI/xAI style cached_input shows here when set.") }} | calls 30d {{ tip("Production llm_calls rows for this model in the last 30 days. The most-used model's count is bolded.") }} | cost 30d {{ tip("Summed cost_usd from production llm_calls in the last 30 days.") }} | tests 30d {{ tip("Admin sandbox experiments + side-by-side evals in the last 30 days.") }} | status {{ tip("active = available for routing. discontinued = archived, not routed. pinned = an env var or DEFAULT_MODEL still points here; archive is blocked.") }} | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| {{ row.company }} | {{ row.model_id }} |
{{ row.version or '—' }} | {% if row.is_frontier %}F{% endif %} {% if row.is_reasoning %}R{% endif %} {% if not row.is_frontier and not row.is_reasoning %}—{% endif %} | {{ price(row.input_price_per_million) }} | {{ price(row.output_price_per_million) }} | {{ price(row.cache_read_price_per_million or row.cached_input_price_per_million) }} | {% if s.calls_30d and s.calls_30d == top_calls %}{{ '{:,}'.format(s.calls_30d) }}{% else %}{{ '{:,}'.format(s.calls_30d) }}{% endif %} | ${{ '%0.2f' % (s.cost_30d_usd | float) }} | {{ '{:,}'.format(s.tests_30d) }} | {% if row.is_discontinued or not row.is_active %} discontinued {% else %} active {% endif %} {% if row.pins %} pinned {% endif %} | {% if is_super_admin %} edit {% else %} view {% endif %} |
F frontier
R reasoning
active available for routing
discontinued archived · no longer routed
prompt -> model{{ tip("Which model each pipeline step runs on. Precedence, highest first: DB assignment (set here) -> per-prompt env var (BROWSER_RECON_LLM_MODEL_) -> global env var (BROWSER_RECON_LLM_MODEL) -> built-in default. A DB assignment always wins; the provenance pill shows which tier is currently in effect so a leftover env var overriding your intent is visible.") }}
{{ assignments | length }} steps
{% if not assignments %}
{% endif %}
// no assignable prompts.
{% else %}
| step {{ tip("The pipeline prompt step. The raw prompt_name (the resolver / assignment key) is shown in muted mono beneath the label.", position="bottom") }} | effective model {{ tip("The model this step resolves to right now across the full cascade, with a pill showing where that choice came from.", position="bottom") }} | {% if is_super_admin %}assign {{ tip("Pick an active registry model to pin this step to (a DB assignment). Choose '— use env / default —' to clear the assignment and fall back to the env/default cascade.", position="bottom") }} | {% endif %}
|---|---|---|
{{ a.label }}
{% if a.legacy %}legacy{% endif %}
{{ a.prompt_name }} |
{{ a.effective_model }}
{{ provenance_pill(a.source) }}
|
{% if is_super_admin %}
{% endif %} |
{{ provenance_pill('assignment') }} set here (DB)
{{ provenance_pill('env_prompt') }} / {{ provenance_pill('env_global') }} from an env var
{{ provenance_pill('default') }} built-in DEFAULT_MODEL
{% if not is_super_admin %}// read-only · super_admin can assign{% endif %}