feat(bothub): add Muse Spark 1.3 Contributor

[4ebuRushka]

Summary

Adds Meta Muse Spark 1.3 Contributor (contributor tier) routed via bothub, as two files:

• \models/meta/muse-spark-1.3-contributor.toml\ — new lab metadata entry
• \providers/bothub/models/muse-spark-1.3-contributor.toml\ — override-only provider file using \�ase_model\

Details

Field	Value
Context / output	1 048 576 / 943 718
Cost (per 1M in/out)	.10 / .20
Modalities (input)	text, image, video, pdf, audio
Reasoning efforts	minimal, low, medium, high, xhigh

• Contributor tier: reduced rate in exchange for allowing Meta to use request data to improve its products (per Meta announcement).
• \max\ effort is not listed: the router rejects it with \
  easoning_effort 'max' is not supported... Supported values: [minimal, low, medium, high, xhigh]\ — Meta holds \max\ back pending safety testing.
• Pricing verified live against the bothub API (\usage.cost: exactly 0.10/0.20 USD per 1M); bothub.ru/models lists 11,79/23,57 ₽ per 1M.
• Reasoning is returned in \message.reasoning\ and mirrored as encrypted
  easoning_details\ → \interleaved = true.
• Tool calling verified working.

Notes / discrepancies

• The existing lab file \models/meta/muse-spark-1.3.toml\ lists \output = 131_072, while OpenRouter/llm-stats/cloudprice report max output 943 718 for 1.3 (and live testing on bothub confirms >131k completion is accepted). I used 943 718 here; happy to align the sibling file in a follow-up if maintainers prefer.
• No \cache_read\ cost: not verifiable on bothub (Meta/OpenRouter list .002, but bothub doesn't surface cached-token pricing).
• No \knowledge\ field: undocumented for this model.

Sources

• https://research.meta.ai/blog/introducing-muse-spark-1-3
• https://openrouter.ai/meta/muse-spark-1.3-contributor
• https://cloudprice.net/models/meta-muse-spark-1-3-contributor
• https://bothub.ru/models (live API verification)

[github-actions]

Action items

• [high] [violation] models/meta/muse-spark-1.3-contributor.toml:1 / providers/bothub/models/muse-spark-1.3-contributor.toml:9 - Check: Contributor tiers must base_model the underlying lab model, not invent a parallel lab identity. Why: Muse Spark 1.3 Contributor is a pricing/data-use tier of Muse Spark 1.3, not a separate lab-built model. First-party Meta (providers/meta/models/muse-spark-1.3-contributor.toml) and other relays (e.g. llmtr 1.2 contributor) use base_model = "meta/muse-spark-1.3" (or …1.2) plus a display name override. There is no existing models/meta/*-contributor.toml. The new lab file duplicates standard 1.3 facts and hard-codes output = 943_718 while models/meta/muse-spark-1.3.toml still has 131_072, splitting one model into two conflicting lab identities. Action: Delete models/meta/muse-spark-1.3-contributor.toml. Point the bothub entry at base_model = "meta/muse-spark-1.3", set name = "Muse Spark 1.3 Contributor", and keep only host deltas (cost, reasoning_options, interleaved, and [limit].output = 943_718 if that max completion is verified on bothub).

[github-actions]

No actionable findings.