Update Kimi K3 reasoning options and add kimi-k3-fast to neuralwatt - #4090
Open
sk0x0y wants to merge 1 commit into
Open
Update Kimi K3 reasoning options and add kimi-k3-fast to neuralwatt#4090sk0x0y wants to merge 1 commit into
sk0x0y wants to merge 1 commit into
Conversation
Neuralwatt now exposes the full K3 reasoning surface: a per-request thinking toggle and graded reasoning effort. The previous toggle-only entry no longer matches the live API. Verified against the live API on 2026-08-05 and aligned with the first-party moonshotai baseline plus ~19 peer relays. - models/moonshotai/kimi-k3.toml: fix base description (toggleable -> configurable low/high/max effort) - providers/neuralwatt/models/kimi-k3.toml: reasoning_options now toggle (chat_template_kwargs.enable_thinking) + effort(low/high/max); drop redundant inherited name. thinking_token_budget is documented but rejected by the current vLLM V2 runner, so it is not declared. - providers/neuralwatt/models/kimi-k3-fast.toml: add non-reasoning variant (reasoning = false, same pricing)
Contributor
|
No actionable findings. |
rekram1-node
reviewed
Aug 6, 2026
| @@ -1,5 +1,5 @@ | |||
| name = "Kimi K3" | |||
| description = "Multimodal Kimi model with 1M context and toggleable max-effort thinking for long-horizon agent work" | |||
| description = "Multimodal Kimi model with 1M context and configurable reasoning effort (low/high/max) for long-horizon agent work" | |||
Collaborator
There was a problem hiding this comment.
this change isnt necessary...
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Updates neuralwatt's Kimi K3 entries to match the current live API surface. The previous entry (
reasoning_options = [{ type = "toggle" }], authored during private beta) predates Neuralwatt exposing graded reasoning effort and thekimi-k3-fastnon-reasoning variant.Host classification
Multi-model relay (
@ai-sdk/openai-compatible; hosts Moonshot / ZAI / DeepSeek / Qwen). Baseline copied from the first-partyproviders/moonshotai/models/kimi-k3.tomlentry plus the ~19 peer relays that all usetoggle + effort(low/high/max).Changes
models/moonshotai/kimi-k3.toml— fix provider-agnostic description ("toggleable max-effort" → configurable low/high/max effort).providers/neuralwatt/models/kimi-k3.toml—reasoning_optionsnowtoggle(chat_template_kwargs.enable_thinking) +effort(low/high/max); drop the redundant inheritedname.providers/neuralwatt/models/kimi-k3-fast.toml— add non-reasoning variant (reasoning = false, same pricing as kimi-k3).Wire fields
chat_template_kwargs: {"enable_thinking": false}reasoning_effort(top-level) =low|high|max(defaultmax)Verification (live API, 2026-08-05)
reasoning_effort=maxreasoning_effort=lowenable_thinking=falsethinking_token_budget=200Note: the Neuralwatt docs prose still states only GLM-5.2 supports
reasoning_effort, but the live/v1/modelsadvertisescapabilities.reasoning_effort: truefor kimi-k3 and the parameter has a measurable effect, so it is declared here. Likewise,thinking_token_budgetis documented generically but is rejected by the current vLLM V2 runner for K3, so it is not declared.Sources
bun validatepasses.