[BOT ISSUE] Together: stale pricing for openai/gpt-oss-20b ($0.075/$0.30 → $0.05/$0.20)

[github-actions]

Gap

The catalog entry for openai/gpt-oss-20b has input_cost_per_mil_tokens: 0.075 and output_cost_per_mil_tokens: 0.3, but Together AI charges $0.05/$0.20 for this model. The entry lists available_providers: ["groq", "together"] — the catalog pricing matches Groq's rate but not Together's.

This is the sibling issue to #988 (openai/gpt-oss-120b stale pricing). The 120b pricing was stale for both Groq and Together; here only Together's pricing diverges.

Official sources

Field	Value	Source	URL
Input/output pricing	$0.05/$0.20	Together AI pricing page	https://www.together.ai/pricing
Input/output pricing	$0.05/$0.20	Together AI serverless models docs	https://docs.together.ai/docs/serverless-models
Input/output pricing	$0.075/$0.30	Groq pricing page	https://groq.com/pricing/

Current vs. correct

Field	Catalog (line 5934)	Together	Groq
input_cost_per_mil_tokens	0.075	0.05 ❌	0.075 ✅
output_cost_per_mil_tokens	0.3	0.20 ❌	0.30 ✅
input_cache_read_cost_per_mil_tokens	0.0375	— (no caching)	0.0375 ✅

Notes

The two providers charge different prices for the same model ID. The catalog currently uses Groq's pricing. Since Together's pricing is lower, users routing through Together would see inflated cost estimates. A few options:

1. Use Together's lower pricing (users on Groq would see underestimated costs)
2. Use Groq's higher pricing (current state — users on Together see overestimated costs)
3. Document the pricing discrepancy or use the lower rate as the default

The Fireworks-specific entry (accounts/fireworks/models/gpt-oss-20b at line 5193) has its own pricing ($0.07/$0.30), showing that provider-specific entries can have independent pricing. The openai/gpt-oss-20b entry is NOT in SYNC_PRESERVED_FIELDS, so its values come from LiteLLM sync.

Comparison with sibling entries

Entry	Input	Output	Providers
accounts/fireworks/models/gpt-oss-20b	0.07	0.30	fireworks
openai/gpt-oss-20b	0.075	0.30	groq, together
openai.gpt-oss-20b-1:0 (Bedrock)	—	—	bedrock

Verification checklist

• Cross-source: Together pricing confirmed on (1) Together AI pricing page ($0.05/$0.20) AND (2) Together AI serverless models docs ($0.05/$0.20)
• Already-fixed check: Entry at line 5934 still shows $0.075/$0.30 on current main
• ID format check: openai/gpt-oss-20b already exists in the catalog
• Duplicate issue check: Issue [BOT ISSUE] Groq/Together: stale pricing for openai/gpt-oss-120b ($0.10/$0.50 → $0.15/$0.60) #988 covers openai/gpt-oss-120b only; no open issue covers the 20b pricing

Local files inspected

• packages/proxy/schema/model_list.json — openai/gpt-oss-20b at line 5934
• packages/proxy/scripts/sync_models.ts — openai/gpt-oss-20b is NOT in SYNC_PRESERVED_FIELDS; accounts/fireworks/models/gpt-oss-20b IS preserved (line 95)

{
  "kind": "stale_metadata",
  "provider": "together",
  "models": ["openai/gpt-oss-20b"],
  "model_specs": {
    "openai/gpt-oss-20b": {
      "format": "openai",
      "flavor": "chat",
      "input_cost_per_mil_tokens": 0.05,
      "output_cost_per_mil_tokens": 0.2,
      "input_cache_read_cost_per_mil_tokens": 0.0375,
      "displayName": "OpenAI GPT-OSS (20B)",
      "reasoning": true,
      "max_input_tokens": 131072,
      "max_output_tokens": 65536,
      "available_providers": ["groq", "together"]
    }
  },
  "source_urls": [
    "https://www.together.ai/pricing",
    "https://docs.together.ai/docs/serverless-models"
  ]
}

[github-actions]

Prepared catalog updates for openai/gpt-oss-20b. Updated models: openai/gpt-oss-20b.

Verification sources: https://www.together.ai/pricing, https://docs.together.ai/docs/serverless-models, https://groq.com/pricing/
Verified metadata applied:

• openai/gpt-oss-20b: format=openai, flavor=chat, displayName=OpenAI GPT-OSS (20B), max_input_tokens=131072, max_output_tokens=65536, pricing(input/output per 1M)=0.05/0.2, input_cache_read_cost_per_mil_tokens=0.0375, providers=groq|together, reasoning=true

Included in the daily batch PR: #1003

[github-actions]

Prepared catalog updates for openai/gpt-oss-20b. Updated models: openai/gpt-oss-20b.

Verification sources: https://www.together.ai/pricing, https://docs.together.ai/docs/serverless-models, https://groq.com/pricing/
Verified metadata applied:

• openai/gpt-oss-20b: format=openai, flavor=chat, displayName=OpenAI GPT-OSS (20B), max_input_tokens=131072, max_output_tokens=65536, pricing(input/output per 1M)=0.05/0.2, input_cache_read_cost_per_mil_tokens=0.0375, providers=groq|together, reasoning=true

Included in the daily batch PR: #1026