feat(zerosignal): add ZeroSignal provider with 12 models

providers/zerosignal/models/glm-5.2.toml

@@ -0,0 +1,25 @@
+# ZeroSignal is a multi-operator relay; operators set their own prices and serving
+# limits. Values below were read from GET /v1/models on the local zs-proxy on 2026-09-16:
+# cost is the lowest advertised operator rate, USD per 1M tokens, inclusive of the
+# protocol fee (what the caller pays); the routed operator may charge differently.
+# Effort values are the network's advertised `allowed_efforts` for this id. Any
+# [limit] override is the advertised context_length / max_completion_tokens.
+# https://docs.zerosignal.ai/for-users/pricing
+# Measured 2026-09-16 (same prompt, temperature 0, max_tokens 4000, reasoning_content
+# length): low and high return identical output (aliased); max is distinct; none
+# still produces a full reasoning trace, so off is not available on this host. The
+# operator advertises the full effort enum, but only high/max are distinct controls.
+# Trace is returned in reasoning_content.
+base_model = "zhipuai/glm-5.2"
+reasoning_options = [{ type = "effort", values = ["high", "max"] }]
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 1.694
+output = 5.324
+cache_read = 0.3146
+
+[limit]
+output = 32_768

providers/zerosignal/models/glm-5.3-flash.toml

@@ -0,0 +1,21 @@
+# ZeroSignal is a multi-operator relay; operators set their own prices and serving
+# limits. Values below were read from GET /v1/models on the local zs-proxy on 2026-09-16:
+# cost is the lowest advertised operator rate, USD per 1M tokens, inclusive of the
+# protocol fee (what the caller pays); the routed operator may charge differently.
+# Effort values are the network's advertised `allowed_efforts` for this id. Any
+# [limit] override is the advertised context_length / max_completion_tokens.
+# https://docs.zerosignal.ai/for-users/pricing
+# Measured 2026-09-16 (same prompt, max_tokens 3000, reasoning tokens from usage):
+# low/high/max accepted with graded traces; none, medium and xhigh are rejected by the
+# host ("always engages in thinking and cannot be disabled"), so off is unavailable.
+# Trace is returned in reasoning_content.
+base_model = "zhipuai/glm-5.3-flash"
+reasoning_options = [{ type = "effort", values = ["low", "high", "max"] }]
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 0.09075
+output = 0.3025
+cache_read = 0.01815

providers/zerosignal/models/glm-5.3.toml

@@ -0,0 +1,24 @@
+# ZeroSignal is a multi-operator relay; operators set their own prices and serving
+# limits. Values below were read from GET /v1/models on the local zs-proxy on 2026-09-16:
+# cost is the lowest advertised operator rate, USD per 1M tokens, inclusive of the
+# protocol fee (what the caller pays); the routed operator may charge differently.
+# Effort values are the network's advertised `allowed_efforts` for this id. Any
+# [limit] override is the advertised context_length / max_completion_tokens.
+# https://docs.zerosignal.ai/for-users/pricing
+# Measured 2026-09-16 (same prompt, max_tokens 3000, reasoning tokens from usage):
+# low/high/max accepted with graded traces; none, medium and xhigh are rejected by the
+# host ("always engages in thinking and cannot be disabled"), so off is unavailable.
+# Trace is returned in reasoning_content.
+base_model = "zhipuai/glm-5.3"
+reasoning_options = [{ type = "effort", values = ["low", "high", "max"] }]
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 1.694
+output = 5.324
+cache_read = 0.3146
+
+[limit]
+output = 32_768

providers/zerosignal/models/google/gemini-3.7-flash.toml

@@ -0,0 +1,19 @@
+# ZeroSignal is a multi-operator relay; operators set their own prices and serving
+# limits. Values below were read from GET /v1/models on the local zs-proxy on 2026-09-16:
+# cost is the lowest advertised operator rate, USD per 1M tokens, inclusive of the
+# protocol fee (what the caller pays); the routed operator may charge differently.
+# Effort values are the network's advertised `allowed_efforts` for this id. Any
+# [limit] override is the advertised context_length / max_completion_tokens.
+# https://docs.zerosignal.ai/for-users/pricing
+# Measured 2026-09-16: none is rejected ("Reasoning is mandatory for this endpoint"),
+# so off is unavailable; low returns zero reasoning tokens, medium/high graded, and
+# xhigh/max are accepted but do not exceed high, so the lab's set is kept. This
+# operator returns the trace in a field named `reasoning`, which is not one of the
+# schema's interleaved field values, so no [interleaved] is declared.
+base_model = "google/gemini-3.7-flash"
+reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }]
+
+[cost]
+input = 1.1
+output = 5.5
+cache_read = 0.11

providers/zerosignal/models/google/gemini-3.8-flash.toml

@@ -0,0 +1,19 @@
+# ZeroSignal is a multi-operator relay; operators set their own prices and serving
+# limits. Values below were read from GET /v1/models on the local zs-proxy on 2026-09-16:
+# cost is the lowest advertised operator rate, USD per 1M tokens, inclusive of the
+# protocol fee (what the caller pays); the routed operator may charge differently.
+# Effort values are the network's advertised `allowed_efforts` for this id. Any
+# [limit] override is the advertised context_length / max_completion_tokens.
+# https://docs.zerosignal.ai/for-users/pricing
+# Measured 2026-09-16: none is rejected ("Reasoning is mandatory for this endpoint"),
+# so off is unavailable; low returns zero reasoning tokens, medium/high graded, and
+# xhigh/max are accepted but do not exceed high, so the lab's set is kept. This
+# operator returns the trace in a field named `reasoning`, which is not one of the
+# schema's interleaved field values, so no [interleaved] is declared.
+base_model = "google/gemini-3.8-flash"
+reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }]
+
+[cost]
+input = 1.1
+output = 5.5
+cache_read = 0.11

providers/zerosignal/models/grok-4.3.toml

@@ -0,0 +1,23 @@
+# ZeroSignal is a multi-operator relay; operators set their own prices and serving
+# limits. Values below were read from GET /v1/models on the local zs-proxy on 2026-09-16:
+# cost is the lowest advertised operator rate, USD per 1M tokens, inclusive of the
+# protocol fee (what the caller pays); the routed operator may charge differently.
+# Effort values are the network's advertised `allowed_efforts` for this id. Any
+# [limit] override is the advertised context_length / max_completion_tokens.
+# https://docs.zerosignal.ai/for-users/pricing
+# Measured 2026-09-16: none returns no reasoning trace (off); graded levels think.
+# Trace is returned in reasoning_content.
+base_model = "xai/grok-4.3"
+reasoning_options = [{ type = "effort", values = ["none", "low", "medium", "high"] }]
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 1.5125
+output = 3.025
+cache_read = 0.242
+
+[limit]
+context = 200_000
+output = 32_768

providers/zerosignal/models/grok-4.5.toml

@@ -0,0 +1,23 @@
+# ZeroSignal is a multi-operator relay; operators set their own prices and serving
+# limits. Values below were read from GET /v1/models on the local zs-proxy on 2026-09-16:
+# cost is the lowest advertised operator rate, USD per 1M tokens, inclusive of the
+# protocol fee (what the caller pays); the routed operator may charge differently.
+# Effort values are the network's advertised `allowed_efforts` for this id. Any
+# [limit] override is the advertised context_length / max_completion_tokens.
+# https://docs.zerosignal.ai/for-users/pricing
+# Measured 2026-09-16: none and max are rejected by the host, so off is unavailable;
+# the graded levels are accepted and all think (token counts within noise of each
+# other, so the lab's set is kept). Trace is returned in reasoning_content.
+base_model = "xai/grok-4.5"
+reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }]
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 2.42
+output = 7.26
+cache_read = 0.363
+
+[limit]
+output = 32_768

providers/zerosignal/models/grok-4.6.toml

@@ -0,0 +1,23 @@
+# ZeroSignal is a multi-operator relay; operators set their own prices and serving
+# limits. Values below were read from GET /v1/models on the local zs-proxy on 2026-09-16:
+# cost is the lowest advertised operator rate, USD per 1M tokens, inclusive of the
+# protocol fee (what the caller pays); the routed operator may charge differently.
+# Effort values are the network's advertised `allowed_efforts` for this id. Any
+# [limit] override is the advertised context_length / max_completion_tokens.
+# https://docs.zerosignal.ai/for-users/pricing
+# Measured 2026-09-16: none and max are rejected by the host, so off is unavailable;
+# the graded levels are accepted and all think (token counts within noise of each
+# other, so the lab's set is kept). Trace is returned in reasoning_content.
+base_model = "xai/grok-4.6"
+reasoning_options = [{ type = "effort", values = ["low", "medium", "high", "xhigh"] }]
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 2.42
+output = 7.26
+cache_read = 0.363
+
+[limit]
+output = 32_768

providers/zerosignal/models/kimi-k3.toml

@@ -0,0 +1,25 @@
+# ZeroSignal is a multi-operator relay; operators set their own prices and serving
+# limits. Values below were read from GET /v1/models on the local zs-proxy on 2026-09-16:
+# cost is the lowest advertised operator rate, USD per 1M tokens, inclusive of the
+# protocol fee (what the caller pays); the routed operator may charge differently.
+# Effort values are the network's advertised `allowed_efforts` for this id. Any
+# [limit] override is the advertised context_length / max_completion_tokens.
+# https://docs.zerosignal.ai/for-users/pricing
+# Measured 2026-09-16: none returns no reasoning trace (off); low/high/max think.
+# Off is effort=none, so no toggle. Moonshot's temperature rule applies through the
+# proxy: thinking modes require temperature 1, none requires 0.6.
+# Trace is returned in reasoning_content.
+base_model = "moonshotai/kimi-k3"
+reasoning_options = [{ type = "effort", values = ["none", "low", "high", "max"] }]
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 3.432
+output = 17.16
+cache_read = 0.3432
+
+[limit]
+context = 1_000_000
+output = 32_768

providers/zerosignal/models/openai/gpt-5.6-luna.toml

@@ -0,0 +1,18 @@
+# ZeroSignal is a multi-operator relay; operators set their own prices and serving
+# limits. Values below were read from GET /v1/models on the local zs-proxy on 2026-09-16:
+# cost is the lowest advertised operator rate, USD per 1M tokens, inclusive of the
+# protocol fee (what the caller pays); the routed operator may charge differently.
+# Effort values are the network's advertised `allowed_efforts` for this id. Any
+# [limit] override is the advertised context_length / max_completion_tokens.
+# https://docs.zerosignal.ai/for-users/pricing
+# Measured 2026-09-16: none returns no reasoning trace (off); all graded levels are
+# accepted by this host, so the set matches the lab entry.
+# This operator returns the trace in a field named `reasoning`, not one of the schema's
+# interleaved field values, so no [interleaved] is declared.
+base_model = "openai/gpt-5.6-luna"
+reasoning_options = [{ type = "effort", values = ["none", "low", "medium", "high", "xhigh", "max"] }]
+
+[cost]
+input = 0.319
+output = 1.936
+cache_read = 0.033