Skip to content

fix(review): put contract-reliable reasoning models first, flagships last #318

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
seonghobae merged 1 commit into from
Jul 5, 2026
+18 −16
Merged

fix(review): put contract-reliable reasoning models first, flagships last #318

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 10 additions & 8 deletions .github/workflows/opencode-review.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2280,14 +2280,16 @@ jobs:
SHARE: "false"
NPM_CONFIG_IGNORE_SCRIPTS: "true"
NO_COLOR: "1"
# Ordered by GitHub Models quota allowance, largest first, so the
# highest-throughput models review before the rate-starved flagships.
# gpt-5/o3 are "Reasoning" tier (8-12 req/day) — putting them first
# (as before) stalled every review until the step timed out, because
# a rate-limited/hung flagship never fell back. Non-reasoning "Low"
# tier (deepseek-v3, mistral, llama-4: 150-450 req/day) and mini
# reasoning models go first; gpt-5 stays last as a quality fallback.
OPENCODE_MODEL_CANDIDATES: "github-models/deepseek/deepseek-v3-0324 github-models/mistral-ai/mistral-medium-2505 github-models/meta/llama-4-maverick-17b-128e-instruct-fp8 github-models/meta/llama-4-scout-17b-16e-instruct github-models/openai/o4-mini github-models/openai/o3-mini github-models/openai/gpt-5-mini github-models/openai/gpt-5-nano github-models/openai/gpt-5-chat github-models/deepseek/deepseek-r1-0528 github-models/deepseek/deepseek-r1 github-models/openai/o3 github-models/openai/gpt-5"
# Ordered contract-reliability first, then quota, with the rate-starved
# flagships last. The mini reasoning models (o4-mini, o3-mini, gpt-5-
# mini/nano/chat) reliably emit the strict review contract — every
# required label and only source-backed findings — so they lead. The
# high-quota non-reasoning models (deepseek-v3, mistral, llama-4) emit
# bare or hallucinated reviews the publish/approve gates reject, so
# they are fallbacks only. gpt-5/o3 ("Reasoning" tier, 8-12 req/day)
# stay last: first-placing them stalled every review until timeout
# because a rate-limited/hung flagship never fell back.
OPENCODE_MODEL_CANDIDATES: "github-models/openai/o4-mini github-models/openai/o3-mini github-models/openai/gpt-5-mini github-models/openai/gpt-5-nano github-models/openai/gpt-5-chat github-models/deepseek/deepseek-r1-0528 github-models/deepseek/deepseek-r1 github-models/deepseek/deepseek-v3-0324 github-models/mistral-ai/mistral-medium-2505 github-models/meta/llama-4-maverick-17b-128e-instruct-fp8 github-models/meta/llama-4-scout-17b-16e-instruct github-models/openai/o3 github-models/openai/gpt-5"
OPENCODE_MODEL_ATTEMPTS: "5"
OPENCODE_RUN_TIMEOUT_SECONDS: "20400"
OPENCODE_EXPORT_TIMEOUT_SECONDS: "120"
Expand Down
16 changes: 8 additions & 8 deletions tests/test_opencode_agent_contract.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -79,9 +79,9 @@ def test_opencode_model_pool_sets_high_effort_for_capable_candidates():
assert candidate_models
assert set(candidate_models).issubset(set(models))
assert candidate_models[:3] == [
"deepseek/deepseek-v3-0324",
"mistral-ai/mistral-medium-2505",
"meta/llama-4-maverick-17b-128e-instruct-fp8",
"openai/o4-mini",
"openai/o3-mini",
"openai/gpt-5-mini",
]
assert {
"openai/gpt-5-chat",
Expand Down Expand Up @@ -286,17 +286,17 @@ def test_workflow_provisions_sandbox_tool_and_reviewer_agent():
assert 'APPROVAL_CHECK_WAIT_ATTEMPTS: "81"' in workflow
assert 'APPROVAL_CHECK_WAIT_SLEEP_SECONDS: "30"' in workflow
assert (
'OPENCODE_MODEL_CANDIDATES: "github-models/deepseek/deepseek-v3-0324 '
"github-models/mistral-ai/mistral-medium-2505 "
"github-models/meta/llama-4-maverick-17b-128e-instruct-fp8 "
"github-models/meta/llama-4-scout-17b-16e-instruct "
"github-models/openai/o4-mini "
'OPENCODE_MODEL_CANDIDATES: "github-models/openai/o4-mini '
"github-models/openai/o3-mini "
"github-models/openai/gpt-5-mini "
"github-models/openai/gpt-5-nano "
'github-models/openai/gpt-5-chat '
"github-models/deepseek/deepseek-r1-0528 "
"github-models/deepseek/deepseek-r1 "
"github-models/deepseek/deepseek-v3-0324 "
"github-models/mistral-ai/mistral-medium-2505 "
"github-models/meta/llama-4-maverick-17b-128e-instruct-fp8 "
"github-models/meta/llama-4-scout-17b-16e-instruct "
"github-models/openai/o3 "
'github-models/openai/gpt-5"'
) in workflow
Expand Down
Loading