feat(bailing): update bailing new models

[lvisei]

Summary

• Add Ling-3.1-flash、Ling-3.0-flash、Ling-3.0-flash-VL
• Remove Ling-1T、Ring-1T Already Deprecated

Pricing

• https://developer.ant-ling.com/en/docs/models/price/#model-pricing

Sources:

• https://developer.ant-ling.com/en/docs/models/price/#model-pricing

[github-actions]

Action items

• [critical] [violation] providers/bailing/models/Ling-3.0-flash.toml:1 - Check: Non-lab hosts must use base_model when the lab model is nameable. Why: models/inclusionai/ling-3.0-flash.toml, models/inclusionai/ling-3.1-flash.toml, and models/inclusionai/ling-3.0-flash-vl.toml already exist and all other hosts (novita-ai, nano-gpt, vercel, llmgateway) use base_model = "inclusionai/...", but this PR adds three full inline definitions under providers/bailing/models/ that duplicate name, description, family, dates, knowledge, open_weights, attachment, tool_call, temperature, [limit]/[modalities]. Action: Rewrite all three files as override-only base_model entries keeping only provider-specific cost, reasoning_options, interleaved, and verified deltas.
• [high] [violation] providers/bailing/models/Ling-3.0-flash.toml:28 - Check: [limit] must be internally consistent and consistent with cited sources. Why: All three new files set context = 256_000 with output = 262144, so output exceeds context; lab entries use context = 262_144 / output = 32_768. Same impossible values appear in Ling-3.1-flash.toml and Ling-3.0-flash-VL.toml. Action: Verify Bailing docs for served context/output, fix values so output <= context, and after base_model keep [limit] only if Bailing truly differs from the lab.
• [high] [possible mistake] providers/bailing/models/Ling-3.0-flash-VL.toml:4 - Check: Display name must distinguish the model ID from the filename. Why: The VL file sets name = "Ling-3.0-flash", identical to providers/bailing/models/Ling-3.0-flash.toml, while the lab entry is Ling 3.0 Flash VL. Action: Verify display name and, after factoring to base_model, inherit or use a distinct VL name.
• [high] [possible mistake] providers/bailing/models/Ling-3.0-flash-VL.toml:9 - Check: attachment/modalities must agree with lab metadata unless the host actually differs. Why: PR sets attachment = false with input = ["text", "image", "video"], but models/inclusionai/ling-3.0-flash-vl.toml sets attachment = true for the same modalities. Action: Verify whether Bailing VL rejects attachments and keep the field only as a real delta from the lab.
• [high] [possible mistake] providers/bailing/models/Ling-3.1-flash.toml:14 - Check: open_weights must reflect publicly available weights. Why: PR sets open_weights = true with no [[weights]], but models/inclusionai/ling-3.1-flash.toml sets open_weights = false with no weights link. Action: Verify weight availability; inherit false via base_model or add a weights URL/license evidence if truly open.
• [medium] [possible mistake] providers/bailing/models/Ling-3.0-flash.toml:22 - Check: All cost values are USD/MTok with conversion noted for non-USD sources. Why: Costs are sourced only to developer.ant-ling.com/en/docs/models/price with no currency/rate/date comment to confirm USD vs CNY conversion for input/output/cache_read on all three models. Action: Confirm values are USD/MTok and add a top-of-file conversion note with rate/date if the source is non-USD.
• [medium] [possible mistake] providers/bailing/models/Ling-1T.toml:1 - Check: Retired provider models should use status = "deprecated", not deletion, without authoritative removal evidence. Why: PR deletes Ling-1T.toml and Ring-1T.toml citing only the current pricing page, while providers/zenmux/models/inclusionai/ling-1t.toml and ring-1t.toml retain the same generation as deprecated. Action: Confirm both IDs are no longer served by the new https://api.ant-ling.com/v1 endpoint and use deprecated instead of deletion if that lifecycle applies.

[github-actions]

Action items

• [high] [violation] providers/bailing/models/Ling-3.1-flash.toml:1 - Check: Non-lab hosts must use base_model when the lab model is nameable. Why: Canonical models/inclusionai/ling-3.1-flash.toml exists and all other hosts (novita-ai, openrouter, vercel, kilo) use base_model = "inclusionai/ling-3.1-flash"; providers/bailing is a different ID from lab inclusionai so full inline is not a first-party-lab exception and duplicates/diverges lab facts. Action: Replace with base_model = "inclusionai/ling-3.1-flash" plus only provider-specific cost, reasoning_options, interleaved and genuine deltas.
• [high] [violation] providers/bailing/models/Ling-3.0-flash.toml:1 - Check: Non-lab hosts must use base_model when the lab model is nameable. Why: Canonical models/inclusionai/ling-3.0-flash.toml exists and peers use base_model; full inline restates family/tool_call/temperature/modalities/limit instead of override-only. Action: Replace with base_model = "inclusionai/ling-3.0-flash" plus only cost, reasoning_options, interleaved and verified deltas.
• [high] [violation] providers/bailing/models/Ling-3.0-flash-VL.toml:1 - Check: Non-lab hosts must use base_model when the lab model is nameable. Why: Canonical models/inclusionai/ling-3.0-flash-vl.toml exists and peers use base_model; full inline duplicates lab metadata with divergent values. Action: Replace with base_model = "inclusionai/ling-3.0-flash-vl" plus only provider-specific fields and verified deltas.
• [medium] [possible mistake] providers/bailing/models/Ling-3.0-flash-VL.toml:4 - Check: Display name must distinguish the variant. Why: Filename is Ling-3.0-flash-VL but name = "Ling-3.0-flash" duplicates the sibling Ling-3.0-flash.toml name; lab names it Ling 3.0 Flash VL. Action: Fix display name to the VL variant.
• [medium] [possible mistake] providers/bailing/models/Ling-3.0-flash-VL.toml:9 - Check: attachment consistency with multimodal input. Why: File sets attachment = false with modalities.input = ["text", "image", "video"], contradicting lab attachment = true and openrouter VL attachment = true for the same model. Action: Verify against Bailing docs; inherit true via base_model unless the host provably rejects attachments.
• [high] [possible mistake] providers/bailing/models/Ling-3.1-flash.toml:14 - Check: open_weights is a lab fact, not a provider override. Why: File claims open_weights = true while canonical models/inclusionai/ling-3.1-flash.toml sets open_weights = false with no weights link (peers inherit false); a host cannot make closed weights open. Action: Verify weights publication; remove the override and inherit via base_model unless public weights are cited.

[github-actions]

Action items

• [critical] [violation] providers/bailing/models/Ling-3.0-flash.toml:1 - Check: Non-lab hosts must use base_model when the lab model is nameable. Why: Canonical lab entries models/inclusionai/ling-3.0-flash.toml, ling-3.1-flash.toml, and ling-3.0-flash-vl.toml already exist and other hosts (novita-ai, nano-gpt, vercel, llmgateway) all use base_model = "inclusionai/..."; this PR adds full inline copies under providers/bailing/ that duplicate and diverge (different description, truncated release_date, different limit, open_weights flip) instead of override-only files. Action: Convert all three new Bailing files to base_model = "inclusionai/ling-..." and keep only provider-specific fields and real deltas (cost, reasoning_options, interleaved, genuine limit/modality overrides).
• [high] [violation] providers/bailing/models/Ling-3.0-flash-VL.toml:4 - Check: Model display name must identify the file's model ID. Why: Both Ling-3.0-flash.toml and Ling-3.0-flash-VL.toml set name = "Ling-3.0-flash", so the VL multimodal file collides with the text-only file and loses the VL identifier. Action: Fix the VL display name (or inherit Ling 3.0 Flash VL via base_model with no local name override unless Bailing documents a different display name).
• [high] [possible mistake] providers/bailing/models/Ling-3.1-flash.toml:14 - Check: open_weights must match lab facts. Why: Lab models/inclusionai/ling-3.1-flash.toml sets open_weights = false with no weights/license, while this PR sets open_weights = true, a direct contradiction for the same underlying model. Action: Verify against first-party weights publication and correct one side; if using base_model, drop the local override unless Bailing documents a different weight release.
• [medium] [possible mistake] providers/bailing/models/Ling-3.0-flash.toml:29 - Check: [limit] overrides must match this host's docs. Why: PR claims context = 256_000 / output = 256_000 (output equals full context) versus lab 262_144 / 32_768 and peer override output = 262_144 on novita-ai; 256_000 vs 262_144 and 256_000 vs 32_768 needs host evidence and looks like a 256k/262144 unit confusion. Action: Cite the exact Bailing limit excerpt for context vs max-output on this API and correct context/output (same check applies to Ling-3.1-flash.toml:30 and Ling-3.0-flash-VL.toml:30).
• [medium] [possible mistake] providers/bailing/models/Ling-3.0-flash.toml:23 - Check: All cost values are USD/MTok with conversion noted when source is non-USD. Why: Files cite developer.ant-ling.com/en/docs/models/price with no currency/rate/date note; if that pricing table is CNY, publishing it as USD understates costs ~7x. Action: Confirm input/output/cache_read are USD/MTok or add a top-of-file conversion comment with rate/date and converted values.