Merge pull request #9 from BaseInfinity/v0.10.3-deepseek-groq-defaults

v0.10.3: DeepSeek + Groq default-model swaps (research-verified)

Author: BaseInfinity

CHANGELOG.md

@@ -2,6 +2,35 @@
 
 All notable changes to opencode-sdlc-wizard.
 
+## [0.10.3] - 2026-05-18
+
+### Changed — two more research-verified default-model bumps
+
+May-17 community-patterns research verified both swaps against live
+provider catalogs; v0.10.3 lands them as a focused patch.
+
+| Provider | Was (v0.10.2) | Now (v0.10.3) | Why |
+|---|---|---|---|
+| `hosted_oss/deepseek` | `deepseek-chat` | `deepseek-v4-flash` | DeepSeek V4 family shipped 2026-04-24; `deepseek-chat` is a moving alias that already points at V4, but explicit pin is clearer for cost/latency-sensitive picks |
+| `hosted_oss/groq` | `llama-3.3-70b-versatile` | `gpt-oss-120b` | Groq's own docs flag `gpt-oss-120b` as "Most Popular for OpenCode"; OpenAI open-weights release on Groq's LPU stack; matches Cerebras default for cross-provider consistency |
+
+PRIVACY.md "Suggested model" table and DeepSeek configure example
+bumped to match.
+
+### Tests
+
+- `test-pick.sh` T12 substring-match for groq updated `/llama/` → `/gpt-oss/`
+- `test-pick.sh` T23 planner-default assertion for groq bumped to `gpt-oss-120b`
+- DeepSeek tests unchanged — `/deepseek/` substring still matches `deepseek-v4-flash`
+- All other suites unchanged (configure-backend tests pin explicit models, not defaults)
+- 357 tests across 12 suites (no count change; T23 / T12 patched in place)
+
+### Compat
+
+- 12 of 14 default-model entries unchanged from v0.10.2.
+- Anyone passing `--model deepseek-chat` explicitly still works (deepseek-chat is still a valid id on DeepSeek's catalog; v0.10.3 just changes what `pick` chooses *by default*).
+- Anyone passing `--model llama-3.3-70b-versatile` explicitly still works (Groq still hosts it; we just default to gpt-oss-120b now per community signal).
+
 ## [0.10.2] - 2026-05-17
 
 ### Added — Planner agent model routing (`--planner-*` flags)

PRIVACY.md

@@ -109,10 +109,10 @@ insufficient and an enterprise tenant is unavailable.
 | Provider | Suggested model | Notes |
 |----------|-----------------|-------|
 | Together | `Qwen/Qwen2.5-Coder-32B-Instruct` | Stable, paid hosting |
-| Groq | `llama-3.3-70b-versatile` | Fastest hosted, free daily quota |
+| Groq | `gpt-oss-120b` | Fastest hosted, free daily quota; OpenAI's open-weights release on Groq's LPU stack |
 | OpenRouter | `qwen/qwen-2.5-coder-32b-instruct` | Aggregator routing across providers |
 | Cerebras | `gpt-oss-120b` or `qwen-3-235b-a22b-instruct-2507` | Free tier, ~2000 tok/s |
-| DeepSeek direct | `deepseek-chat` | Cheapest paid hosted (~$0.14/M cache-miss) |
+| DeepSeek direct | `deepseek-v4-flash` | Cheapest paid hosted; V4 family shipped 2026-04-24 |
 | NVIDIA NIM (`nvidia_nim`) | `deepseek-ai/deepseek-r1` | Free credits at build.nvidia.com |
 
 ```bash
@@ -126,7 +126,7 @@ bash .opencode/scripts/configure-backend.sh \
 export DEEPSEEK_API_KEY="..."
 bash .opencode/scripts/configure-backend.sh \
      --tier hosted_oss --provider deepseek \
-     --model "deepseek-chat"
+     --model "deepseek-v4-flash"
 
 # Stable paid — Together
 export TOGETHER_API_KEY="..."

package.json

@@ -1,6 +1,6 @@
 {
   "name": "opencode-sdlc-wizard",
-  "version": "0.10.2",
+  "version": "0.10.3",
   "description": "SDLC enforcement for OpenCode CLI — privacy-first, any-backend portability with a four-tier backend picker plus an OSS-tier cross-model-review skill so the full SDLC loop can run with zero Anthropic+OpenAI lock-in. Ships JSON Schemas for review artifacts so any consumer (cross-model-review, ditto, CI) can validate. Install with `npx opencode-sdlc-wizard init`. Sibling of agentic-sdlc-wizard and codex-sdlc-wizard.",
   "bin": {
     "opencode-sdlc-wizard": "cli/bin/opencode-sdlc-wizard.js"

scripts/pick-backend.sh

@@ -164,10 +164,10 @@ default_model_for() {
                                   echo "anthropic.claude-sonnet-4-5-20250929-v1:0" ;;
     hosted_oss/together|hosted_oss/togetherai)
                                   echo "Qwen/Qwen2.5-Coder-32B-Instruct" ;;
-    hosted_oss/groq)              echo "llama-3.3-70b-versatile" ;;
+    hosted_oss/groq)              echo "gpt-oss-120b" ;;
     hosted_oss/openrouter)        echo "qwen/qwen-2.5-coder-32b-instruct" ;;
     hosted_oss/cerebras)          echo "gpt-oss-120b" ;;
-    hosted_oss/deepseek)          echo "deepseek-chat" ;;
+    hosted_oss/deepseek)          echo "deepseek-v4-flash" ;;
     hosted_oss/nvidia_nim|hosted_oss/nvidia|hosted_oss/nvidia-nim)
                                   echo "deepseek-ai/deepseek-r1" ;;
     proprietary/anthropic)        echo "claude-opus-4-7" ;;

tests/test-pick.sh

@@ -238,7 +238,7 @@ for combo in \
     "private_local/llama_cpp:qwen" \
     "private_local/vllm:Qwen" \
     "hosted_oss/together:Qwen" \
-    "hosted_oss/groq:llama" \
+    "hosted_oss/groq:gpt-oss" \
     "hosted_oss/openrouter:qwen" \
     "hosted_oss/cerebras:gpt-oss" \
     "hosted_oss/deepseek:deepseek" \
@@ -414,8 +414,8 @@ if [ -x "$SCRIPT" ]; then
   if [ -f "$CONFIGURE_STUB_LOG" ] \
      && grep -q -- "--planner-tier hosted_oss" "$CONFIGURE_STUB_LOG" \
      && grep -q -- "--planner-provider groq" "$CONFIGURE_STUB_LOG" \
-     && grep -q -- "--planner-model llama-3.3-70b-versatile" "$CONFIGURE_STUB_LOG"; then
-    pass "Planner: --planner-provider groq → planner-model llama-3.3-70b-versatile default"
+     && grep -q -- "--planner-model gpt-oss-120b" "$CONFIGURE_STUB_LOG"; then
+    pass "Planner: --planner-provider groq → planner-model gpt-oss-120b default (v0.10.3)"
   else
     fail "Planner default not forwarded"
     cat "$CONFIGURE_STUB_LOG" 2>/dev/null | head -3 >&2 || true