Conversation
Contributor
Action items
|
Author
|
All five findings were fair. The
Cells are reasoning tokens. Per item:
The other seven ids advertise sets that already match their lab entries and are unchanged. PR body updated with the same table. |
Contributor
Action items
|
…rleaved where observed
Author
|
Both items addressed. Remaining seven ids measured. Same method as before (same prompt,
|
Contributor
|
No actionable findings. |
github-actions
Bot
added
the
reviewer: ready
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds ZeroSignal as an OpenAI-compatible provider. ZeroSignal is a pay-per-use inference network; models are served by independent operators and reached through a local proxy (
zs-proxy) athttp://127.0.0.1:9376/v1, the same shape as theatomic-chatandlmstudioentries. The proxy does not check the API key (admission comes from the user's wallet), soZEROSIGNAL_API_KEYcan be any non-empty value.ZeroSignal is a multi-model relay, not a lab, so every entry uses
base_modeland is override-only. I'm on the ZeroSignal team.Scope
12 models, all with existing lab metadata under
models/. The network serves more (about 50 ids), but I've limited this PR to the ones where the network advertises an explicit effort set that lines up with the lab entry. The rest (ids with no advertised reasoning controls, ids from operators advertising a genericlow/medium/highon every model, and a few operator fine-tunes with no lab entry) I'd rather verify on the wire before publishing controls for them. Separate PR if/when that's done.Data provenance
Everything below was read from
GET /v1/modelson the local proxy on 2026-09-16 (the proxy aggregates the on-chain operator registry). It is a local endpoint, so the relevant entries are inlined under the fold rather than linked.Cost is the lowest advertised operator rate for the id, USD per 1M tokens, inclusive of the protocol fee (what the caller pays). Operators price independently, so this is a floor; the routed operator may charge more.
cache_readis included only where an operator actually discounts it belowinput; there is no cache-write premium on the network, socache_writeis omitted.reasoning_optionswere measured on the host for all 12 ids rather than taken from the network's advertisedallowed_efforts(that field is declared by each operator in node config). Method: same prompt,max_tokens3000 to 4000 so nothing truncates, temperature 0 unless the model dictates one, comparing reasoning tokens fromusageand the returned trace across every effort value, 2026-09-16 against the local proxy. Where the measured controls differ from the lab entry, the file's leading comment says what was observed.glm-5.3reasoning_contentglm-5.3-flashreasoning_contentglm-5.2reasoning_contentkimi-k3reasoning_contentgrok-4.6reasoning_contentgrok-4.5reasoning_contentgrok-4.3reasoning_contentgoogle/gemini-3.7-flashreasoninggoogle/gemini-3.8-flashreasoningopenai/gpt-5.6-lunareasoningopenai/gpt-5.6-terraopenai/gpt-6-astraCells are reasoning tokens; "rejected" is a host error for that value; blank means not probed. Notes:
glm-5.2returned byte-identical output forlowandhighand a full trace fornone, so onlyhigh/max(Z.ai's own set) are authored and off is unavailable.kimi-k3also enforces Moonshot's temperature rule through the proxy (1 for thinking levels, 0.6 fornone). Grok 4.5/4.6 graded counts are within noise, so xAI's sets are kept. Gemini acceptsxhigh/maxbut they do not exceedhigh, so Google's set is kept;lowyields zero reasoning tokens.openai/gpt-6-astracould not be exercised beyondnonebecause every graded call was rate-limited upstream on the day; its authored set is the operator's advertised one, identical to OpenAI's entry, and the comment says so.[interleaved]is declared asfield = "reasoning_content"on the seven ids that returned the trace in that field. The two Geminis andgpt-5.6-lunareturn it in a field namedreasoning, which is not one of the schema's values;gpt-5.6-terrareturned it in one measurement and not another;gpt-6-astrawas not observed. Those five carry no[interleaved]and a comment explaining why.[limit]overrides are only where the operator's advertisedcontext_lengthormax_completion_tokensdiffers from the lab file. Where the network advertises no output cap, the lab value is inherited rather than guessed.glm-5.3zhipuai/glm-5.3glm-5.3-flashzhipuai/glm-5.3-flashglm-5.2zhipuai/glm-5.2kimi-k3moonshotai/kimi-k3grok-4.6xai/grok-4.6grok-4.5xai/grok-4.5grok-4.3xai/grok-4.3google/gemini-3.7-flashgoogle/gemini-3.7-flashgoogle/gemini-3.8-flashgoogle/gemini-3.8-flashopenai/gpt-5.6-lunaopenai/gpt-5.6-lunaopenai/gpt-5.6-terraopenai/gpt-5.6-terraopenai/gpt-6-astraopenai/gpt-6-astraRaw
/v1/modelsentries for the 12 ids (2026-09-16)[ { "id": "glm-5.3", "object": "model", "created": 0, "owned_by": "zerosignal", "canonical_slug": "zai/glm-5.3", "context_length": 1000000, "pricing": { "prompt": "0.000001694", "completion": "0.000005324", "input_cache_read": "0.0000003146" }, "top_provider": { "context_length": 1000000, "max_completion_tokens": 32768 }, "reasoning": { "supported": true, "allowed_efforts": [ "low", "high", "max" ] }, "tool_use": true }, { "id": "glm-5.3-flash", "object": "model", "created": 0, "owned_by": "zerosignal", "canonical_slug": "zai/glm-5.3-flash", "context_length": 1000000, "architecture": { "input_modalities": [ "text", "image", "video" ], "output_modalities": [ "text" ] }, "pricing": { "prompt": "0.00000009075", "completion": "0.0000003025", "input_cache_read": "0.00000001815" }, "top_provider": { "context_length": 1000000, "max_completion_tokens": 131072 }, "reasoning": { "supported": true, "allowed_efforts": [ "low", "high", "max" ] }, "tool_use": true }, { "id": "glm-5.2", "object": "model", "created": 0, "owned_by": "zerosignal", "hugging_face_id": "zai-org/GLM-5.2", "context_length": 1000000, "pricing": { "prompt": "0.000001694", "completion": "0.000005324", "input_cache_read": "0.0000003146" }, "top_provider": { "context_length": 1000000, "max_completion_tokens": 32768 }, "reasoning": { "supported": true, "allowed_efforts": [ "none", "minimal", "low", "medium", "high", "xhigh", "max" ] }, "tool_use": true }, { "id": "kimi-k3", "object": "model", "created": 0, "owned_by": "zerosignal", "hugging_face_id": "moonshotai/Kimi-K3", "context_length": 1000000, "architecture": { "input_modalities": [ "text", "image", "video" ] }, "pricing": { "prompt": "0.000003432", "completion": "0.00001716", "input_cache_read": "0.0000003432" }, "top_provider": { "context_length": 1000000, "max_completion_tokens": 32768 }, "reasoning": { "supported": true, "allowed_efforts": [ "low", "high", "max" ] }, "tool_use": true }, { "id": "grok-4.6", "object": "model", "created": 0, "owned_by": "zerosignal", "canonical_slug": "xai/grok-4.6", "context_length": 500000, "architecture": { "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ] }, "pricing": { "prompt": "0.00000242", "completion": "0.00000726", "input_cache_read": "0.000000363" }, "top_provider": { "context_length": 500000, "max_completion_tokens": 32768 }, "reasoning": { "supported": true, "allowed_efforts": [ "low", "medium", "high", "xhigh" ] }, "tool_use": true }, { "id": "grok-4.5", "object": "model", "created": 0, "owned_by": "zerosignal", "canonical_slug": "xai/grok-4.5", "context_length": 500000, "architecture": { "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ] }, "pricing": { "prompt": "0.00000242", "completion": "0.00000726", "input_cache_read": "0.000000363" }, "top_provider": { "context_length": 500000, "max_completion_tokens": 32768 }, "reasoning": { "supported": true, "allowed_efforts": [ "low", "medium", "high" ] }, "tool_use": true }, { "id": "grok-4.3", "object": "model", "created": 0, "owned_by": "zerosignal", "canonical_slug": "xai/grok-4.3", "context_length": 200000, "architecture": { "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ] }, "pricing": { "prompt": "0.0000015125", "completion": "0.000003025", "input_cache_read": "0.000000242" }, "top_provider": { "context_length": 200000, "max_completion_tokens": 32768 }, "reasoning": { "supported": true, "allowed_efforts": [ "low", "medium", "high" ] }, "tool_use": true }, { "id": "google/gemini-3.7-flash", "object": "model", "created": 0, "owned_by": "zerosignal", "canonical_slug": "google/gemini-3.7-flash", "context_length": 1048576, "architecture": { "input_modalities": [ "text", "image", "video", "audio" ], "output_modalities": [ "text" ] }, "pricing": { "prompt": "0.0000011", "completion": "0.0000055", "input_cache_read": "0.00000011" }, "top_provider": { "context_length": 1048576, "max_completion_tokens": 65536 }, "reasoning": { "supported": true, "allowed_efforts": [ "low", "medium", "high" ] }, "tool_use": true }, { "id": "google/gemini-3.8-flash", "object": "model", "created": 0, "owned_by": "zerosignal", "canonical_slug": "google/gemini-3.8-flash", "context_length": 1048576, "architecture": { "input_modalities": [ "text", "image", "video", "audio" ], "output_modalities": [ "text" ] }, "pricing": { "prompt": "0.0000011", "completion": "0.0000055", "input_cache_read": "0.00000011" }, "top_provider": { "context_length": 1048576, "max_completion_tokens": 65536 }, "reasoning": { "supported": true, "allowed_efforts": [ "low", "medium", "high" ] }, "tool_use": true }, { "id": "openai/gpt-5.6-luna", "object": "model", "created": 0, "owned_by": "zerosignal", "hugging_face_id": "openai/gpt-5.6-luna", "context_length": 1050000, "architecture": { "input_modalities": [ "text", "image", "file" ], "output_modalities": [ "text" ] }, "pricing": { "prompt": "0.000000319", "completion": "0.000001936", "input_cache_read": "0.000000033" }, "top_provider": { "context_length": 1050000, "max_completion_tokens": 128000 }, "reasoning": { "supported": true, "allowed_efforts": [ "low", "medium", "high", "xhigh", "max" ] }, "tool_use": true }, { "id": "openai/gpt-5.6-terra", "object": "model", "created": 0, "owned_by": "zerosignal", "hugging_face_id": "openai/gpt-5.6-terra", "context_length": 1050000, "architecture": { "input_modalities": [ "text", "image", "file" ], "output_modalities": [ "text" ] }, "pricing": { "prompt": "0.000002662", "completion": "0.000015972", "input_cache_read": "0.0000002662" }, "top_provider": { "context_length": 1050000, "max_completion_tokens": 128000 }, "reasoning": { "supported": true, "allowed_efforts": [ "low", "medium", "high", "xhigh", "max" ] }, "tool_use": true }, { "id": "openai/gpt-6-astra", "object": "model", "created": 0, "owned_by": "zerosignal", "hugging_face_id": "openai/gpt-6-astra", "context_length": 1050000, "architecture": { "input_modalities": [ "text", "image", "file" ], "output_modalities": [ "text" ] }, "pricing": { "prompt": "0.0000121", "completion": "0.0000605", "input_cache_read": "0.00000121" }, "top_provider": { "context_length": 1050000, "max_completion_tokens": 128000 }, "reasoning": { "supported": true, "allowed_efforts": [ "low", "medium", "high", "xhigh", "max" ] }, "tool_use": true } ]Verification
bun validatepasses locally.packages/weband ranOPENCODE_MODELS_PATH="dist/_api.json" opencode; ZeroSignal appears in/connectand the 12 models in/models, and a completion against the running proxy streams.logo.svgis a single-colorcurrentColormark with a square viewBox and no fixed size.