Skip to content

Repository files navigation

Opencode Modelselect

GitHub release GitHub last commit GitHub issues GitHub pull requests GitHub license

Model routing ecosystem for OpenCode: GitHub Action for CI, live plugin for sessions, OpenChamber status view, and LiveBench-driven model ranking from one central config (data/model-config.json), so model updates propagate with no sync and no edits.

Components

Component Path Deep docs
Select Model Action github-action/ github-action/README.md
Modelselect Plugin plugin/ plugin/README.md
OpenChamber Status View openchamber-modelselect/ openchamber-modelselect/README.md
  • Select Model Action preselects the model in CI before running the OpenCode step (task-type + tier → model output). The root action.yml is a thin shim pointing at github-action/src/index.js, so uses: dianlight/opencode-modelselect-action@v1 keeps working; the subpath .../github-action@v1 resolves the same code directly. See the component README for inputs, outputs, tier: auto probing, max-cost budgets and fail-closed behavior.
  • Modelselect Plugin routes the model live inside OpenCode sessions, inferring the task-type from project signals, prompt text and agent tag. OpenCode v2 only. See the component README for install (v2/OpenChamber), options, host detection, the modelselect/auto virtual model, the /modelselect command, token resolution and trial runs.
  • OpenChamber Status View shows the plugin's per-session pick plus the global on/off/auto mode switch in the Work Status panel, and records session activity into the host-detection session map the plugin reads. See the component README for install, build and test.

Quick start

- name: Select model
  id: resolve
  uses: dianlight/opencode-modelselect-action@v1
  with:
    task-type: review
    tier: go

- uses: anomalyco/opencode/github@<sha>
  with:
    model: ${{ steps.resolve.outputs.model }}

There are no default models: if the config is unreachable or the entry is missing, the step fails hard unless an explicit fallback-model is given.

Central config and maintenance

data/model-config.json is the actual configuration, keyed by task-type only (task-types.<name>.{go,free}), consumed by every component at startup. The maintenance script (scripts/opencode_maintenance.py, run by the opencode-maintenance workflow) recomputes the config each run and commits it directly; task types are defined in config/task-types.yaml. To change a model: run mise run maintenance and review the committed data/model-config.json diff — downstream workflows pick it up automatically at their next run.

Model Recommendations by Task Type

Automatically updated by opencode-maintenance workflow. Last updated: 2026-09-30 02:42 UTC. LiveBench data: 353 models scored. LiveBench snapshot: 2026_06_25. Source: https://livebench.ai/table_2026_06_25.csv Free-first threshold: 5%. Blended cost weights: 75% in / 25% out ($/1M). Costs shown as in/out $/1M (Free = $0).

Task Type Description Best Zen Best Free Best Go
plan (Plan) Planning, architecture decisions, task decomposition opencode/gpt-5-codex (96.1, $1.07/$8.5) opencode/muse-spark-1.3-contributor-free (90.9, Free) 🏆 opencode-go/muse-spark-1.3-contributor-free (90.9, Free)
issue-triage (Issue Triage) Triage, label, categorize, route issues opencode/gpt-5.5-pro (81.6, $30/$180) opencode/longcat-2.5-preview-free (86.0, Free) 🏆 opencode-go/longcat-2.5-preview-free (86.0, Free)
review (Review) Review PRs, pull requests, diffs opencode/gpt-5-codex (96.1, $1.07/$8.5) opencode/muse-spark-1.3-contributor-free (90.9, Free) 🏆 opencode-go/muse-spark-1.3-contributor-free (90.9, Free)
ui-design (UI Design) UI design, components, layouts, mockups opencode/gpt-6-astra (86.3, $10/$50) opencode/muse-spark-1.3-contributor-free (68.4, Free) 🏆 opencode-go/hy4-preview (84.0, $0.834/$2.501)
ui-testing (UI Testing) Playwright, Cypress, E2E, frontend tests opencode/claude-opus-5-5 (78.7, $4/$20) opencode/mimo-v2.6-flash-free (71.0, Free) 🏆 opencode-go/deepseek-v4.1-flash (78.4, $0.3/$1.2)
api-testing (API Testing) API testing, integration tests, OpenAPI, Postman opencode/claude-opus-5-5 (78.7, $4/$20) opencode/mimo-v2.6-flash-free (71.0, Free) 🏆 opencode-go/deepseek-v4.1-flash (78.4, $0.3/$1.2)
docs (Docs) Documentation, READMEs, changelogs, docstrings opencode/gpt-5.5-pro (81.6, $30/$180) opencode/longcat-2.5-preview-free (86.0, Free) 🏆 opencode-go/longcat-2.5-preview-free (86.0, Free)
debug (Debug) Debugging, reproductions, crash and exception triage opencode/gpt-5-codex (96.1, $1.07/$8.5) opencode/muse-spark-1.3-contributor-free (90.9, Free) 🏆 opencode-go/muse-spark-1.3-contributor-free (90.9, Free)
refactor (Refactor) Refactoring, cleanup, tech-debt reduction opencode/claude-opus-5-5 (78.7, $4/$20) opencode/mimo-v2.6-flash-free (71.0, Free) 🏆 opencode-go/deepseek-v4.1-flash (78.4, $0.3/$1.2)
security (Security) Security review, vulnerabilities, hardening opencode/gpt-5-codex (96.1, $1.07/$8.5) opencode/muse-spark-1.3-contributor-free (90.9, Free) 🏆 opencode-go/muse-spark-1.3-contributor-free (90.9, Free)
code (Code) Code generation, implementation, features opencode/claude-opus-5-5 (78.7, $4/$20) opencode/mimo-v2.6-flash-free (71.0, Free) 🏆 opencode-go/deepseek-v4.1-flash (78.4, $0.3/$1.2)
mechanical-engineer (Mechanical Engineer) Mechanical engineering, calculations, CAD, thermodynamics, materials opencode/gpt-5-codex (96.1, $1.07/$8.5) opencode/muse-spark-1.3-contributor-free (90.9, Free) 🏆 opencode-go/muse-spark-1.3-contributor-free (90.9, Free)
web-search (Web Search) Ricerche approfondite sul web, sintesi multi-fonte opencode/gpt-5-codex (96.1, $1.07/$8.5) opencode/muse-spark-1.3-contributor-free (90.9, Free) 🏆 opencode-go/muse-spark-1.3-contributor-free (90.9, Free)
generic (Generic) General Q&A, explanations, analysis, everything else opencode/gpt-5.5-pro (81.6, $30/$180) opencode/longcat-2.5-preview-free (86.0, Free) 🏆 opencode-go/longcat-2.5-preview-free (86.0, Free)
small-model (Small Model) Lightweight utility tasks: commit messages, session titles/renames, summaries opencode/gpt-5.5-pro (81.6, $30/$180) opencode/longcat-2.5-preview-free (86.0, Free) 🏆 opencode-go/longcat-2.5-preview-free (86.0, Free)

LiveBench Score Reference

Token costs ($/1M, blended 75% in / 25% out). Value = Overall ÷ Blended $/1M (higher = better value). Value shows — when it cannot be computed: Free models cost $0 (value would be infinite), and paid models with unknown pricing (— in the cost columns) have no divisor.

Model Tier Source Best For Overall Coding Reasoning Vision Instruction Following In $/1M Out $/1M Blended $/1M Value
big-pickle Free 📋 Fallback UITest, APITest 61.5 67.0 61.5 8.0 60.0 Free Free Free —
deepseek-flash Go (Paid) 📋 Fallback Plan, Review 67.3 69.2 89.9 5.0 80.4 — — — —
deepseek-v4-flash Go (Paid) ✅ LiveBench Plan, Review 66.1 50.3 73.1 46.9 69.4 $0.14 $0.28 $0.175 377.7
deepseek-v4-flash-free Free ✅ LiveBench Plan, Review 66.1 50.3 73.1 46.9 69.4 Free Free Free —
deepseek-v4-flash-vision-exp Go (Paid) ✅ LiveBench Plan, Review 77.7 66.3 86.1 57.7 75.2 $0.14 $0.28 $0.175 444.0
deepseek-v4-pro Go (Paid) ✅ LiveBench Plan, Review 72.6 53.6 83.9 56.4 70.3 $1.74 $3.48 $2.175 33.4
deepseek-v4.1-flash Go (Paid) ✅ LiveBench Plan, Review 81.4 78.4 88.7 61.6 74.0 $0.3 $1.2 $0.525 155.0
glm-5 Go (Paid) ✅ LiveBench Plan, Review 68.7 62.5 74.0 63.6 65.0 $1 $3.2 $1.55 44.3
glm-5.1 Go (Paid) ✅ LiveBench Plan, Review 70.6 63.1 75.6 60.3 69.5 $1.4 $4.4 $2.15 32.8
glm-5.2 Go (Paid) ✅ LiveBench Plan, Review 73.4 62.9 82.1 60.7 68.2 $1.4 $4.4 $2.15 34.1
glm-5.3 Go (Paid) ✅ LiveBench Plan, Review 76.6 68.1 84.4 60.2 72.3 $1.4 $4.4 $2.15 35.6
glm-5.3-flash Go (Paid) ✅ LiveBench Plan, Review 71.1 65.6 78.2 60.3 64.5 $0.15 $0.5 $0.2375 299.4
gpt-5.6-luna Go (Paid) ✅ LiveBench Plan, Review 73.7 62.2 84.8 51.2 66.6 $0.2 $1.2 $0.45 163.8
gpt-6-luna Go (Paid) ✅ LiveBench Plan, Review 72.2 62.3 82.7 57.5 63.7 $0.1 $0.5 $0.2 361.0
grok-4.5 Go (Paid) ✅ LiveBench Plan, Review 77.1 61.3 86.9 66.4 73.9 $2 $6 $3 25.7
grok-4.6 Go (Paid) ✅ LiveBench Plan, Review 79.0 64.9 89.1 67.1 74.3 $2 $6 $3 26.3
grok-4.7 Go (Paid) ✅ LiveBench Plan, Review 78.3 63.3 87.9 72.4 74.3 $2 $6 $3 26.1
hy3 Go (Paid) 📋 Fallback Plan, Review 54.0 55.0 60.0 8.0 58.0 $0.14 $0.58 $0.25 216.0
hy3-preview Go (Paid) 📋 Fallback Plan, Review 54.0 55.0 60.0 8.0 58.0 — — — —
hy4-preview Go (Paid) 📋 Fallback Plan, Review 79.2 68.9 85.0 84.0 78.0 $0.834 $2.501 $1.2508 63.3
jev-1.13-free Free 📋 Fallback Plan, Review 67.8 30.0 67.8 5.0 50.0 Free Free Free —
kimi-k2.5 Go (Paid) ✅ LiveBench Plan, Review 69.2 60.1 76.7 55.0 65.3 $0.6 $3 $1.2 57.7
kimi-k2.6 Go (Paid) ✅ LiveBench Plan, Review 70.9 59.6 77.9 58.1 69.7 $0.95 $4 $1.7125 41.4
kimi-k2.7-code Go (Paid) ✅ LiveBench Plan, Review 68.8 57.0 76.9 55.7 65.9 $0.95 $4 $1.7125 40.2
kimi-k3 Go (Paid) ✅ LiveBench Plan, Review 79.5 69.9 87.0 72.6 74.9 $3 $15 $6 13.2
ling-3.0-flash-fin-free Free 📋 Fallback Triage, Docs 45.0 48.0 48.0 5.0 50.0 Free Free Free —
longcat-2.0 Go (Paid) 📋 Fallback Triage, Docs 60.0 70.0 89.0 5.0 90.0 $0.3 $1.2 $0.525 114.3
longcat-2.5-preview-free Free 📋 Fallback Triage, Docs 55.0 66.0 84.0 40.0 86.0 Free Free Free —
mimo-v2-omni Go (Paid) 📋 Fallback Design, Triage 50.0 42.0 48.0 55.0 50.0 — — — —
mimo-v2-pro Go (Paid) ✅ LiveBench Plan, Review 58.4 45.5 65.8 43.6 57.8 — — — —
mimo-v2.5 Go (Paid) 📋 Fallback Triage, Docs 62.0 60.0 64.0 58.0 65.0 $0.14 $0.28 $0.175 354.3
mimo-v2.5-free Free 📋 Fallback Triage, Docs 58.0 56.0 60.0 54.0 62.0 Free Free Free —
mimo-v2.5-pro Go (Paid) 📋 Fallback Triage, Docs 68.0 66.0 70.0 5.0 74.0 $0.435 $0.87 $0.5437 125.1
mimo-v2.6-flash Go (Paid) 📋 Fallback Plan, Review 64.0 75.0 84.0 40.0 76.0 $0.14 $0.28 $0.175 365.7
mimo-v2.6-flash-free Free 📋 Fallback Plan, Review 60.0 71.0 80.0 38.0 72.0 Free Free Free —
mimo-v2.6-pro Go (Paid) 📋 Fallback Plan, Review 70.0 80.0 88.0 50.0 82.0 $0.435 $0.87 $0.5437 128.7
minimax-m2.5 Go (Paid) ✅ LiveBench Plan, Review 60.3 59.3 62.3 31.3 62.2 $0.3 $1.2 $0.525 114.9
minimax-m2.7 Go (Paid) ✅ LiveBench Plan, Review 65.0 52.0 72.4 34.0 67.4 $0.3 $1.2 $0.525 123.8
minimax-m3 Go (Paid) ✅ LiveBench Plan, Review 67.5 51.7 76.7 50.2 66.9 $0.3 $1.2 $0.525 128.6
muse-spark-1.2-contributor Go (Paid) ✅ LiveBench Plan, Review 78.9 65.6 88.7 61.7 74.9 $0.1 $0.2 $0.125 631.2
muse-spark-1.2-contributor-free Free ✅ LiveBench Plan, Review 78.9 65.6 88.7 61.7 74.9 Free Free Free —
muse-spark-1.3-contributor Go (Paid) ✅ LiveBench Plan, Review 82.4 70.9 90.9 68.4 78.9 $0.1 $0.2 $0.125 659.2
muse-spark-1.3-contributor-free Free ✅ LiveBench Plan, Review 82.4 70.9 90.9 68.4 78.9 Free Free Free —
nemotron-3-ultra-free Free ✅ LiveBench Plan, Review 68.7 51.5 75.6 47.8 74.0 Free Free Free —
nemotron-3.5-lightning-free Free 📋 Fallback Triage, Docs 45.0 36.2 58.0 5.0 72.0 Free Free Free —
omen-alpha Go (Paid) 📋 Fallback Plan, Review 56.0 57.9 65.0 40.0 62.0 — — — —
qwen3.5-plus Go (Paid) 📋 Fallback Plan, Review 58.0 52.0 62.0 42.0 60.0 $0.2 $1.2 $0.45 128.9
qwen3.6-plus Go (Paid) ✅ LiveBench Plan, Review 69.0 56.1 77.0 52.5 67.7 $0.5 $3 $1.125 61.3
qwen3.7-max Go (Paid) ✅ LiveBench Plan, Review 74.1 55.8 82.4 58.7 76.6 $2.5 $7.5 $3.75 19.8
qwen3.7-plus Go (Paid) 📋 Fallback Plan, Triage 66.0 62.0 72.0 62.0 72.0 $0.4 $1.6 $0.7 94.3
qwen3.8-flash Go (Paid) ✅ LiveBench Plan, Review 77.3 66.0 84.4 55.8 77.2 $0.15 $0.47 $0.23 336.1
qwen3.8-max Go (Paid) ✅ LiveBench Plan, Review 79.5 67.9 87.9 58.6 77.2 $2 $6 $3 26.5
space-bunny-free Free 📋 Fallback Plan, Review 30.0 28.0 40.0 38.0 35.0 Free Free Free —

Sponsor

Github Sponsor Buy Me a Coffee

Referral

If you're interested in subscribing to an OpenCode Go plan, click this referral link — you'll get a $5 credit, and $5 will be donated to support this project.

License

MIT License

About

Model routing ecosystem for OpenCode: GitHub Action for CI, live plugin for sessions, OpenChamber status view, and LiveBench-driven model ranking from one central config.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages