Bug: Should use only free models when no providers configured

[cloudwaddie-agent]

Bug Description

When oh-my-opencode is installed/configured and the user selects no providers, subagents should ONLY use free models. Currently, they incorrectly inherit or use a paid model (glm-4.7) instead of defaulting to free models only.

Expected Behavior

When no providers are configured:

• Subagents should use ONLY free models (e.g., opencode/big-pickle, opencode/kimi-k2.5-free, opencode/minimax-m2.5-free)
• Should NOT inherit the parent model's configuration
• Should default to free models since no paid providers were set up

Actual Behavior

• User configures: no providers selected (free models only)
• Main agent may work correctly with a free model
• Subagent (child agent) uses glm-4.7 (a paid model), causing errors since GLM 4.7 is no longer free
• Subagent should instead use only free models when no providers are configured

Steps to Reproduce

1. Configure oh-my-opencode with no providers (free models only)
2. Set the main model to a free model like opencode/kimi-k2.5-free
3. Spawn a subagent (task with category or subagent_type)
4. Observe that the subagent attempts to use glm-4.7 instead of a free model

Environment

• oh-my-opencode version: latest
• Configuration: No providers selected (free models only)

Additional Context

Reported in: CloudWaddie/actions-agent#67

The fix should ensure that when no providers are configured, the system defaults to using ONLY free models, rather than inheriting any parent model configuration or falling back to a paid model.

[cbrunnkvist]

Absolutely. This is what it currently produces if you say no to all the providers:

▲  No model providers configured. Using opencode/glm-4.7-free as fallback.

The resulting JSON:

{
  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
  "agents": {
    "hephaestus": {
      "model": "opencode/glm-4.7-free"
    },
    "oracle": {
      "model": "opencode/glm-4.7-free"
    },
    "librarian": {
      "model": "opencode/glm-4.7-free"
    },
    "explore": {
      "model": "opencode/glm-4.7-free"
    },
    "multimodal-looker": {
      "model": "opencode/glm-4.7-free"
    },
    "prometheus": {
      "model": "opencode/glm-4.7-free"
    },
    "metis": {
      "model": "opencode/glm-4.7-free"
    },
    "momus": {
      "model": "opencode/glm-4.7-free"
    },
    "atlas": {
      "model": "opencode/glm-4.7-free"
    }
  },
  "categories": {
    "visual-engineering": {
      "model": "opencode/glm-4.7-free"
    },
    "ultrabrain": {
      "model": "opencode/glm-4.7-free"
    },
    "deep": {
      "model": "opencode/glm-4.7-free"
    },
    "artistry": {
      "model": "opencode/glm-4.7-free"
    },
    "quick": {
      "model": "opencode/glm-4.7-free"
    },
    "unspecified-low": {
      "model": "opencode/glm-4.7-free"
    },
    "unspecified-high": {
      "model": "opencode/glm-4.7-free"
    },
    "writing": {
      "model": "opencode/glm-4.7-free"
    }
  }
}

Interestingly, despite me deleting ~/.config/opencode/oh-my-opencode.json prior to npx oh-my-opencode install, I still see

● Existing configuration detected: Claude=max20, Gemini=no

[cbrunnkvist]

FWIW IMHO IME this is, as of 2026-02-16, the ultimate freeloader setup for OmO (ought to be the default when the user tells the setup wizard that they don't want to use any of the suggested providers):

{
  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
  "agents": {
    "sisyphus": {"model": "opencode/minimax-m2.5-free"},
    "hephaestus": {"model": "opencode/kimi-k2.5-free"},
    "oracle": {"model": "opencode/big-pickle", "variant": "high"},
    "librarian": {"model": "opencode/minimax-m2.5-free"},
    "explore": {"model": "opencode/kimi-k2.5-free"},
    "multimodal-looker": {"model": "opencode/minimax-m2.5-free"},
    "prometheus": {"model": "opencode/big-pickle"},
    "metis": {"model": "opencode/kimi-k2.5-free"},
    "momus": {"model": "opencode/minimax-m2.5-free"},
    "atlas": {"model": "opencode/kimi-k2.5-free"},
    "frontend-ui-ux": {"model": "opencode/minimax-m2.5-free"}
  },
  "categories": {
    "visual-engineering": {"model": "opencode/minimax-m2.5-free"},
    "ultrabrain": {"model": "opencode/big-pickle", "variant": "xhigh"},
    "deep": {"model": "opencode/kimi-k2.5-free"},
    "artistry": {"model": "opencode/big-pickle"},
    "quick": {"model": "opencode/kimi-k2.5-free"},
    "unspecified-low": {"model": "opencode/kimi-k2.5-free"},
    "unspecified-high": {"model": "opencode/minimax-m2.5-free"},
    "writing": {"model": "opencode/minimax-m2.5-free"}
  }
}

Allocation strategy:

Role	Model	Why
Premium tasks (oracle, prometheus, ultrabrain, artistry)	big-pickle	Best quality for critical reasoning
Heavy lifting (sisyphus, categories)	minimax-m2.5-free	Quality 41.97 - best available
Light/fast tasks (explore, quick, metis)	kimi-k2.5-free	Quick turnaround

In general, you're actually in pretty decent shape. minimax-m2.5-free at should be comparable to models costing $3-6/M tokens. The main sacrifice is speed - free models may be slower, and the Kimi free in particular is prone to 24h-window rate limiting, but, hey, it's pretty awesome.

[moinulmoin]

no actually, it shouldnt be this way, that means not only sacrificing the speed but also quality, and also free models are tempo, will keep changing

[CloudWaddie]

FWIW IMHO IME this is, as of 2026-02-16, the ultimate freeloader setup for OmO (ought to be the default when the user tells the setup wizard that they don't want to use any of the suggested providers):

{
"$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
"agents": {
"sisyphus": {"model": "opencode/minimax-m2.5-free"},
"hephaestus": {"model": "opencode/kimi-k2.5-free"},
"oracle": {"model": "opencode/big-pickle", "variant": "high"},
"librarian": {"model": "opencode/minimax-m2.5-free"},
"explore": {"model": "opencode/kimi-k2.5-free"},
"multimodal-looker": {"model": "opencode/minimax-m2.5-free"},
"prometheus": {"model": "opencode/big-pickle"},
"metis": {"model": "opencode/kimi-k2.5-free"},
"momus": {"model": "opencode/minimax-m2.5-free"},
"atlas": {"model": "opencode/kimi-k2.5-free"},
"frontend-ui-ux": {"model": "opencode/minimax-m2.5-free"}
},
"categories": {
"visual-engineering": {"model": "opencode/minimax-m2.5-free"},
"ultrabrain": {"model": "opencode/big-pickle", "variant": "xhigh"},
"deep": {"model": "opencode/kimi-k2.5-free"},
"artistry": {"model": "opencode/big-pickle"},
"quick": {"model": "opencode/kimi-k2.5-free"},
"unspecified-low": {"model": "opencode/kimi-k2.5-free"},
"unspecified-high": {"model": "opencode/minimax-m2.5-free"},
"writing": {"model": "opencode/minimax-m2.5-free"}
}
}

Allocation strategy:
Role Model Why
Premium tasks (oracle, prometheus, ultrabrain, artistry) big-pickle Best quality for critical reasoning
Heavy lifting (sisyphus, categories) minimax-m2.5-free Quality 41.97 - best available
Light/fast tasks (explore, quick, metis) kimi-k2.5-free Quick turnaround

In general, you're actually in pretty decent shape. minimax-m2.5-free at should be comparable to models costing $3-6/M tokens. The main sacrifice is speed - free models may be slower, and the Kimi free in particular is prone to 24h-window rate limiting, but, hey, it's pretty awesome.

i agree this should be the default. whilst models do change everything about opencode changes during development, so oh my opencode should keep up

[code-yeongyu]

Triage Analysis

After reviewing the codebase, I need to clarify the actual behavior:

Current Behavior

When no providers are configured, the system correctly uses free models only:

• ULTIMATE_FALLBACK = "opencode/glm-4.7-free" (src/cli/model-fallback.ts:21)
• ALL agents and categories are assigned this free model (verified by test snapshot)

The Issue Description Appears Incorrect

The bug report states "subagent uses glm-4.7 (a paid model)" - but the code explicitly uses glm-4.7-free, not the paid version.

What This Actually Is

This is a feature request for diversified free model assignments, not a bug:

• Current: All agents use the same free model (glm-4.7-free)
• Requested: Different free models per agent (e.g., kimi-k2.5-free, minimax-m2.5-free, big-pickle)

Recommendation

1. If the claim is that paid models are being used: Please provide evidence (logs/config showing glm-4.7 without the -free suffix)
2. If requesting diversified free models: This is a valid enhancement but should be tracked as a feature request, not a bug
3. Note on maintenance: Free models change frequently, making diversified assignments high-maintenance

@cloudwaddie-agent Can you confirm whether you're seeing actual errors with paid models, or if this is a request for the diversified setup suggested by @cbrunnkvist?

[cbrunnkvist]

The claim (by me & @cloudwaddie-agent) is that non-existing models are being wrongly assigned when you don't have any paid provider i.e. you answer "No" to all the setup questions.

The claim is accurate as of ${now}: a model named opencode/glm-4.7-free does not exist (anymore). There is one called opencode/glm-4.7, which is NOT free and thus should not be a fallback either, when the user has just confirmed that no paid providers are to be used.

HOWEVER what might be confusing people is that OpenCode still LISTS a "free" tier GLM on their UI, but as shown empirically above, it is not available despite this:

I hope this helps settle the confusion :-)

Evidence

Using "first time launch" = no providers configured at all:

[MacBookAir ~/Developer/www]
% T=$(mktemp -d) ; export XDG_CONFIG_HOME=$T/config XDG_DATA_HOME=$T/data
[MacBookAir ~/Developer/www]
% opencode auth list
Performing one time database migration, may take a few minutes...
Database migration complete.

┌  Credentials /var/folders/br/5_pjgzf533s9334_n3bws8w00000gn/T/tmp.OWt0gjXAPq/data/opencode/auth.json
│
└  0 credentials

[MacBookAir ~/Developer/www]
% opencode models
opencode/big-pickle
opencode/gpt-5-nano
opencode/kimi-k2.5-free
opencode/minimax-m2.5-free
[MacBookAir ~/Developer/www]
% opencode --version
1.2.6

Once "Connect -> OpenCode Zen -> API Key has been set up, we have a GLM variant, but still no -free (ref):

[MacBookAir ~/Developer/www]
% unset XDG_CONFIG_HOME XDG_DATA_HOME
[MacBookAir ~/Developer/www]
% opencode auth list

┌  Credentials ~/.local/share/opencode/auth.json
│
●  OpenCode Zen api
│
└  1 credentials

[MacBookAir ~/Developer/www]
% opencode models | wc -l
      60
[MacBookAir ~/Developer/www]
% opencode models | grep glm-4.7
opencode/glm-4.7
[MacBookAir ~/Developer/www]
% opencode --version
1.2.6

[CloudWaddie]

@code-yeongyu glm-4.7-free is no longer a model. it was retired and now is only accessible on a pay as you go plan. all references to glm-4.7-free should be changed to another model, e.g. kimi k2.5, minimax m2.5, or big pickle.

[CloudWaddie]

i am the maintainer of cloudwaddie-agent btw

[deepfriedmind]

opencode/glm-4.7-free should be replaced with opencode/glm-5-free.

[cbrunnkvist]

opencode/glm-5-free has only been available since a few hours back and is said to be free "this week", no idea if that means from this week of until the end of this week or 7 day from now.

It seems glitchy, fluctuating between english and something else as a response to "Hello.", and it's about 5x slower than kimi/minmax currently:

Run 1:
opencode/glm-5-free | YES | Hello! 😊 I’m your GLM model trained by Z.ai. How | 12249ms

Run 2:
opencode/glm-5-free | YES | 你好！我是Z.ai训练的GLM大语言模型。很高兴能与你交流。 有什么我能帮助你的问题或者你想了解的 | 11304ms

Run 3:
opencode/glm-5-free | YES | Hello! 😊 I’m your GLM model trained by Z.ai. How | 12249ms

[wommy]

1. i feel like glm-5 more aligns with codex big n slow type workloads
2. big-pickle resolved to minimax-m2.5-free now, its still the only model with a reasoning level, or so i believe
3. https://opencode.ai/zen/v1/models and https://github.com/anomalyco/opencode/blob/dev/packages/web/src/content/docs/zen.mdx#pricing i think is the only way to decipher the free models - can anyone else get gpt-5-nano to do tool calls / writes or am i missing something
4. kilo has pretty good free models too https://api.kilo.ai/api/gateway/models follow their blogpost to set it up

• more minimax/minimax-m2.5:free z-ai/glm-5:free arcee-ai/trinity-large-preview:free
• giga-potato, which supposedly has vision, but i cant get it to work, might be on my end
• just came out with x-ai/grok-code-fast-1:optimized:free
• openrouter/aurora-alpha which kinda feels kinda like a grok-fast type
• openrouter/free which just picks a random free one, could be a nice backup, tho ive always found its better to hand pick them myself
• stepfun/step-3.5-flash:free i think this one really needs to be dialed in parameter wise to be useful
• openrouter/bodybuilder can be used to change 2 together, might be useful for some of the sisyphus routing stuff under the hood
• corethink:free another one i cant get to work

i think the value of the kilo api is specifically bc we dont have to just only use big-pickle to turn the knobs on thinking level, i thing thats a big limitation of opencode at the current moment

from their api

{
  [
    {
      "id": "minimax/minimax-m2.5:free",
      "canonical_slug": "minimax/minimax-m2.5:free",
      "hugging_face_id": "",
      "name": "MiniMax: MiniMax M2.5 (free)",
      "created": 1756238927,
      "description": "MiniMax-M2.5 is a SOTA large language model designed for real-world productivity. Trained in a diverse range of complex real-world digital working environments, M2.5 builds upon the coding expertise of M2.1 to extend into general office work, reaching fluency in generating and operating Word, Excel, and Powerpoint files, context switching between diverse software environments, and working across different agent and human teams. Scoring 80.2% on SWE-Bench Verified, 51.3% on Multi-SWE-Bench, and 76.3% on BrowseComp, M2.5 is also more token efficient than previous generations, having been trained to optimize its actions and output through planning.",
      "context_length": 204800,
      "architecture": {
        "modality": "text-\u003Etext",
        "input_modalities": [
          "text"
        ],
        "output_modalities": [
          "text"
        ],
        "tokenizer": "Other",
        "instruct_type": null
      },
      "pricing": {
        "prompt": "0.0000000",
        "completion": "0.0000000",
        "request": "0",
        "image": "0",
        "web_search": "0",
        "internal_reasoning": "0",
        "input_cache_read": "0.00000000"
      },
      "top_provider": {
        "context_length": 204800,
        "max_completion_tokens": 131072,
        "is_moderated": false
      },
      "per_request_limits": null,
      "supported_parameters": [
        "max_tokens",
        "temperature",
        "tools",
        "reasoning",
        "include_reasoning"
      ],
      "default_parameters": {

      },
      "preferredIndex": 1,
      "settings": {
        "included_tools": [
          "search_and_replace"
        ],
        "excluded_tools": [
          "apply_diff",
          "edit_file"
        ]
      }
    },
    {
      "id": "z-ai/glm-5:free",
      "canonical_slug": "z-ai/glm-5:free",
      "hugging_face_id": "",
      "name": "Z.ai: GLM 5 (free)",
      "created": 1756238927,
      "description": "GLM-5 is Z.ai’s flagship open-source foundation model engineered for complex systems design and long-horizon agent workflows. Built for expert developers, it delivers production-grade performance on large-scale programming tasks, rivaling leading closed-source models. With advanced agentic planning, deep backend reasoning, and iterative self-correction, GLM-5 moves beyond code generation to full-system construction and autonomous execution.",
      "context_length": 202800,
      "architecture": {
        "modality": "text-\u003Etext",
        "input_modalities": [
          "text"
        ],
        "output_modalities": [
          "text"
        ],
        "tokenizer": "Other",
        "instruct_type": null
      },
      "pricing": {
        "prompt": "0.0000000",
        "completion": "0.0000000",
        "request": "0",
        "image": "0",
        "web_search": "0",
        "internal_reasoning": "0",
        "input_cache_read": "0.00000000"
      },
      "top_provider": {
        "context_length": 202800,
        "max_completion_tokens": 131072,
        "is_moderated": false
      },
      "per_request_limits": null,
      "supported_parameters": [
        "max_tokens",
        "temperature",
        "tools",
        "reasoning",
        "include_reasoning"
      ],
      "default_parameters": {

      },
      "preferredIndex": 2,
      "versioned_settings": {
        "4.146.0": {
          "included_tools": [
            "write_file",
            "edit_file"
          ],
          "excluded_tools": [
            "apply_diff"
          ]
        }
      }
    },
    {
      "id": "giga-potato",
      "canonical_slug": "giga-potato",
      "hugging_face_id": "",
      "name": "Giga Potato (free)",
      "created": 1756238927,
      "description": "Giga Potato is a stealth model deeply optimized for agentic programming, with visual understanding capability. It is provided free of charge in Kilo Code for a limited time.\n**Note:** Prompts and completions are logged and may be used to improve the model.",
      "context_length": 256000,
      "architecture": {
        "modality": "text+image-\u003Etext",
        "input_modalities": [
          "text",
          "image"
        ],
        "output_modalities": [
          "text"
        ],
        "tokenizer": "Other",
        "instruct_type": null
      },
      "pricing": {
        "prompt": "0.0000000",
        "completion": "0.0000000",
        "request": "0",
        "image": "0",
        "web_search": "0",
        "internal_reasoning": "0",
        "input_cache_read": "0.00000000"
      },
      "top_provider": {
        "context_length": 256000,
        "max_completion_tokens": 32000,
        "is_moderated": false
      },
      "per_request_limits": null,
      "supported_parameters": [
        "max_tokens",
        "temperature",
        "tools",
        "reasoning",
        "include_reasoning"
      ],
      "default_parameters": {

      },
      "preferredIndex": 3,
      "versioned_settings": {
        "4.146.0": {
          "included_tools": [
            "write_file",
            "edit_file"
          ],
          "excluded_tools": [
            "apply_diff"
          ]
        }
      }
    },
    {
      "id": "arcee-ai/trinity-large-preview:free",
      "name": "Arcee AI: Trinity Large Preview (free)",
      "created": 1769552670,
      "description": "Trinity-Large-Preview is a frontier-scale open-weight language model from Arcee, built as a 400B-parameter sparse Mixture-of-Experts with 13B active parameters per token using 4-of-256 expert routing. \n\nIt excels in creative writing, storytelling, role-play, chat scenarios, and real-time voice assistance, better than your average reasoning model usually can. But we’re also introducing some of our newer agentic performance. It was trained to navigate well in agent harnesses like OpenCode, Cline, and Kilo Code, and to handle complex toolchains and long, constraint-filled prompts. \n\nThe architecture natively supports very long context windows up to 512k tokens, with the Preview API currently served at 128k context using 8-bit quantization for practical deployment. Trinity-Large-Preview reflects Arcee’s efficiency-first design philosophy, offering a production-oriented frontier model with open weights and permissive licensing suitable for real-world applications and experimentation.",
      "architecture": {
        "input_modalities": [
          "text"
        ],
        "output_modalities": [
          "text"
        ],
        "tokenizer": "Other"
      },
      "top_provider": {
        "is_moderated": false,
        "context_length": 131000,
        "max_completion_tokens": null
      },
      "pricing": {
        "prompt": "0",
        "completion": "0",
        "image": "0",
        "request": "0",
        "web_search": "0",
        "internal_reasoning": "0"
      },
      "context_length": 131000,
      "per_request_limits": null,
      "supported_parameters": [
        "max_tokens",
        "response_format",
        "structured_outputs",
        "temperature",
        "tools",
        "top_k",
        "top_p"
      ],
      "preferredIndex": 4
    },
    {
      "id": "x-ai/grok-code-fast-1:optimized:free",
      "canonical_slug": "x-ai/grok-code-fast-1:optimized:free",
      "hugging_face_id": "",
      "name": "xAI: Grok Code Fast 1 Optimized (free)",
      "created": 1756238927,
      "description": "An optimized variant of Grok Code Fast 1, provided free of charge for a limited time. **Note:** All prompts and completions for this model are logged by the provider and may be used to improve their services.",
      "context_length": 256000,
      "architecture": {
        "modality": "text-\u003Etext",
        "input_modalities": [
          "text"
        ],
        "output_modalities": [
          "text"
        ],
        "tokenizer": "Other",
        "instruct_type": null
      },
      "pricing": {
        "prompt": "0.0000000",
        "completion": "0.0000000",
        "request": "0",
        "image": "0",
        "web_search": "0",
        "internal_reasoning": "0",
        "input_cache_read": "0.00000000"
      },
      "top_provider": {
        "context_length": 256000,
        "max_completion_tokens": 10000,
        "is_moderated": false
      },
      "per_request_limits": null,
      "supported_parameters": [
        "max_tokens",
        "temperature",
        "tools",
        "reasoning",
        "include_reasoning"
      ],
      "default_parameters": {

      },
      "preferredIndex": 13
    },
    {
      "id": "openrouter/aurora-alpha",
      "name": "Aurora Alpha (free)",
      "created": 1770611225,
      "description": "This is a cloaked model provided to the community to gather feedback. A reasoning model designed for speed. It is built for coding assistants, real-time conversational applications, and agentic workflows.\n\nDefault reasoning effort is set to medium for fast responses. For agentic coding use cases, we recommend changing effort to high. \n\nNote: All prompts and completions for this model are logged by the provider and may be used to improve the model.",
      "architecture": {
        "input_modalities": [
          "text"
        ],
        "output_modalities": [
          "text"
        ],
        "tokenizer": "Other"
      },
      "top_provider": {
        "is_moderated": false,
        "context_length": 128000,
        "max_completion_tokens": 50000
      },
      "pricing": {
        "prompt": "0",
        "completion": "0",
        "image": "0",
        "request": "0",
        "web_search": "0",
        "internal_reasoning": "0"
      },
      "context_length": 128000,
      "per_request_limits": null,
      "supported_parameters": [
        "include_reasoning",
        "max_tokens",
        "reasoning",
        "reasoning_effort",
        "response_format",
        "structured_outputs",
        "temperature",
        "tool_choice",
        "tools"
      ]
    },
    {
      "id": "openrouter/free",
      "name": "Free Models Router (free)",
      "created": 1769917427,
      "description": "The simplest way to get free inference. openrouter/free is a router that selects free models at random from the models available on OpenRouter. The router smartly filters for models that support features needed for your request such as image understanding, tool calling, structured outputs and more. ",
      "architecture": {
        "input_modalities": [
          "text",
          "image"
        ],
        "output_modalities": [
          "text"
        ],
        "tokenizer": "Router"
      },
      "top_provider": {
        "is_moderated": false,
        "context_length": null,
        "max_completion_tokens": null
      },
      "pricing": {
        "prompt": "0",
        "completion": "0"
      },
      "context_length": 200000,
      "per_request_limits": null,
      "supported_parameters": [
        "frequency_penalty",
        "include_reasoning",
        "max_tokens",
        "min_p",
        "presence_penalty",
        "reasoning",
        "repetition_penalty",
        "response_format",
        "seed",
        "stop",
        "structured_outputs",
        "temperature",
        "tool_choice",
        "tools",
        "top_k",
        "top_p"
      ]
    },
    {
      "id": "stepfun/step-3.5-flash:free",
      "name": "StepFun: Step 3.5 Flash (free)",
      "created": 1769728337,
      "description": "Step 3.5 Flash is StepFun's most capable open-source foundation model. Built on a sparse Mixture of Experts (MoE) architecture, it selectively activates only 11B of its 196B parameters per token. It is a reasoning model that is incredibly speed efficient even at long contexts.",
      "architecture": {
        "input_modalities": [
          "text"
        ],
        "output_modalities": [
          "text"
        ],
        "tokenizer": "Other"
      },
      "top_provider": {
        "is_moderated": false,
        "context_length": 256000,
        "max_completion_tokens": 256000
      },
      "pricing": {
        "prompt": "0",
        "completion": "0"
      },
      "context_length": 256000,
      "per_request_limits": null,
      "supported_parameters": [
        "frequency_penalty",
        "include_reasoning",
        "max_tokens",
        "reasoning",
        "stop",
        "temperature",
        "tools",
        "top_p"
      ]
    },
    {
      "id": "openrouter/bodybuilder",
      "name": "Body Builder (beta)",
      "created": 1764903653,
      "description": "Transform your natural language requests into structured OpenRouter API request objects. Describe what you want to accomplish with AI models, and Body Builder will construct the appropriate API calls. Example: \"count to 10 using gemini and opus.\"\n\nThis is useful for creating multi-model requests, custom model routers, or programmatic generation of API calls from human descriptions.\n\n**BETA NOTICE**: Body Builder is in beta, and currently free. Pricing and functionality may change in the future.",
      "architecture": {
        "input_modalities": [
          "text"
        ],
        "output_modalities": [
          "text"
        ],
        "tokenizer": "Router"
      },
      "top_provider": {
        "is_moderated": false,
        "context_length": null,
        "max_completion_tokens": null
      },
      "pricing": {
        "prompt": "-1",
        "completion": "-1"
      },
      "context_length": 128000,
      "per_request_limits": null,
      "supported_parameters": []
    },
    {
      "id": "corethink:free",
      "canonical_slug": "corethink:free",
      "hugging_face_id": "",
      "name": "CoreThink (free)",
      "created": 1756238927,
      "description": "CoreThink - AI that reasons through problems instead of guessing. Available free of charge in Kilo for a limited time.",
      "context_length": 78000,
      "architecture": {
        "modality": "text-\u003Etext",
        "input_modalities": [
          "text"
        ],
        "output_modalities": [
          "text"
        ],
        "tokenizer": "Other",
        "instruct_type": null
      },
      "pricing": {
        "prompt": "0.0000000",
        "completion": "0.0000000",
        "request": "0",
        "image": "0",
        "web_search": "0",
        "internal_reasoning": "0"
      },
      "top_provider": {
        "context_length": 78000,
        "max_completion_tokens": 8192,
        "is_moderated": false
      },
      "per_request_limits": null,
      "supported_parameters": [
        "max_tokens",
        "temperature",
        "tools"
      ],
      "default_parameters": {

      }
    }
  ]
}

i think its really important to dial in the parameters for each of them

[cbrunnkvist]

This ticket is not about finding the most perfect setup by using other-than-the premium models & -providers.

It is about resolving the issue that OmO does not work if you answer "no" to all of its provider setup questions. We must work within constraints, or tickets will never get resolved.

I suggest moving the "what is the ultimate freeloader / budget setup to include in the Guidelines as of ${date}" into a separate ticket.

[CloudWaddie]

@code-yeongyu my suggestion is to dynamically resolve the free models using the opencode models endpoint or something like that.