Resolve dev conflicts in #5563

.github/workflows/ci-fixer.yml

@@ -2,7 +2,7 @@ name: Dev CI Fixer
 
 on:
   workflow_run:
-    workflows: [Deploy]
+    workflows: [Deploy, Sync Model Catalogs]
     types: [completed]
   workflow_dispatch:
 
@@ -160,7 +160,7 @@ jobs:
           Failed log excerpt:
           EOF
             cat "$LOG_FILE"
-          } | opencode run --agent ci-fixer -m opencode/grok-4.5 | tee "$RESPONSE_FILE"
+          } | opencode run --agent ci-fixer -m opencode/muse-spark-1.3 --variant xhigh | tee "$RESPONSE_FILE"
 
       - name: Check changed paths
         if: steps.budget.outputs.run == 'true' && steps.budget-cache.outputs.cache-hit != 'true'

.github/workflows/issue-fixer.yml

@@ -60,7 +60,7 @@ jobs:
             + "If it is a feature request, a request to track a new kind of information, a question, or any miscellaneous non-catalog-data request, do not edit files. Respond briefly that it needs maintainer review and no automated fix was opened."
           ' "$ISSUE_FILE" > "$PROMPT_FILE"
 
-          opencode run --agent issue-fixer -m opencode/grok-4.5 --format json < "$PROMPT_FILE" | tee "$EVENTS_FILE"
+          opencode run --agent issue-fixer -m opencode/muse-spark-1.3 --variant xhigh --format json < "$PROMPT_FILE" | tee "$EVENTS_FILE"
 
           if ! jq -ers 'map(select(.type == "text") | .part.text) | last | select(length > 0)' "$EVENTS_FILE" > "$RESPONSE_FILE"; then
             echo "Issue fixer did not produce a final response." >&2

.github/workflows/opencode.yml

@@ -27,4 +27,5 @@ jobs:
         env:
           OPENCODE_API_KEY: ${{ secrets.OPENCODE_API_KEY }}
         with:
-          model: opencode/grok-4.5
+          model: opencode/muse-spark-1.3
+          variant: xhigh

.github/workflows/pr-reviewer.yml

@@ -78,7 +78,7 @@ jobs:
           export PR_REVIEW_READY_FILE
           rm -f "$PR_REVIEW_READY_FILE"
 
-          opencode run --agent pr-reviewer -m opencode/grok-4.5 --format json <<'EOF' | tee "$EVENTS_FILE"
+          opencode run --agent pr-reviewer -m opencode/muse-spark-1.3 --variant xhigh --format json <<'EOF' | tee "$EVENTS_FILE"
           Review this pull request using the trusted reviewer instructions. Start with `.pr-review/pull-request.json`, `.pr-review/diff.patch`, `AGENTS.md`, and the contributing guidance in `README.md`. Read `sync.md`, the reasoning-options audit guide, schema code, and nearby base-revision files when relevant to the changed files. Use only the read, glob, grep, and mark-pr-ready tools. Return only the final review comment in the agent's required output format. Never include progress narration or passed-check summaries.
           EOF
 

.github/workflows/retry-issue-fixer.yml

@@ -0,0 +1,102 @@
+name: Retry stalled issue fixes
+
+on:
+  schedule:
+    - cron: "41 */6 * * *"
+  workflow_dispatch:
+
+permissions:
+  contents: write
+  issues: read
+  pull-requests: read
+
+concurrency:
+  group: retry-stalled-issue-fixes
+  cancel-in-progress: false
+
+jobs:
+  retry:
+    if: github.repository == 'anomalyco/models.dev'
+    runs-on: ubuntu-latest
+    steps:
+      - name: Redispatch oldest issues without pull requests
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          set -euo pipefail
+
+          cutoff="$(date -u -d '1 hour ago' '+%Y-%m-%dT%H:%M:%SZ')"
+          owner="${GITHUB_REPOSITORY%/*}"
+          repo="${GITHUB_REPOSITORY#*/}"
+
+          issues="$({
+            gh issue list \
+              --repo "$GITHUB_REPOSITORY" \
+              --state open \
+              --label automation \
+              --label model-sync \
+              --label missing-model \
+              --limit 1000 \
+              --json number,createdAt,labels
+          } | jq --arg cutoff "$cutoff" '
+            map(select(.createdAt <= $cutoff)) | sort_by(.createdAt)
+          ')"
+
+          dispatched=0
+          dispatched_issues=()
+          while IFS= read -r issue; do
+            number="$(jq -r '.number' <<< "$issue")"
+            provider="$(jq -r '
+              [.labels[].name | select(startswith("provider:")) | ltrimstr("provider:")]
+              | first // empty
+            ' <<< "$issue")"
+
+            if [ -z "$provider" ] || [ "$provider" = "openai" ]; then
+              continue
+            fi
+
+            closing_prs="$(gh api graphql \
+              -F owner="$owner" \
+              -F repo="$repo" \
+              -F number="$number" \
+              -f query='query($owner: String!, $repo: String!, $number: Int!) {
+                repository(owner: $owner, name: $repo) {
+                  issue(number: $number) {
+                    closedByPullRequestsReferences(first: 1, includeClosedPrs: true) {
+                      totalCount
+                    }
+                  }
+                }
+              }' \
+              --jq '.data.repository.issue.closedByPullRequestsReferences.totalCount')"
+
+            if [ "$closing_prs" -gt 0 ]; then
+              continue
+            fi
+
+            jq -n \
+              --arg provider "$provider" \
+              --argjson issue_number "$number" \
+              '{
+                event_type: "missing-model",
+                client_payload: {
+                  provider: $provider,
+                  issue_number: $issue_number
+                }
+              }' \
+              | gh api "repos/$GITHUB_REPOSITORY/dispatches" --method POST --input -
+
+            dispatched=$((dispatched + 1))
+            dispatched_issues+=("#$number")
+            if [ "$dispatched" -eq 3 ]; then
+              break
+            fi
+          done < <(jq -c '.[]' <<< "$issues")
+
+          if [ "$dispatched" -eq 0 ]; then
+            echo "No stalled automated issues were eligible for retry." | tee -a "$GITHUB_STEP_SUMMARY"
+          else
+            printf 'Redispatched %d issue(s): %s\n' \
+              "$dispatched" "${dispatched_issues[*]}" \
+              | tee -a "$GITHUB_STEP_SUMMARY"
+          fi

.github/workflows/sync-models.yml

@@ -77,6 +77,7 @@ jobs:
           DEEPINFRA_API_KEY: ${{ secrets.DEEPINFRA_API_KEY }}
           DIGITALOCEAN_API_TOKEN: ${{ secrets.DIGITALOCEAN_API_TOKEN }}
           DIGITALOCEAN_ACCESS_TOKEN: ${{ secrets.DIGITALOCEAN_ACCESS_TOKEN }}
+          FIREWORKS_API_KEY: ${{ secrets.FIREWORKS_API_KEY }}
           HF_TOKEN: ${{ secrets.HF_TOKEN }}
           OPENROUTER_API_KEY: ${{ secrets.OPENROUTER_API_KEY }}
           OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
@@ -88,6 +89,8 @@ jobs:
           GEMINI_API_KEY: ${{ secrets.GEMINI_API_KEY }}
           GOOGLE_GENERATIVE_AI_API_KEY: ${{ secrets.GOOGLE_GENERATIVE_AI_API_KEY }}
           XAI_API_KEY: ${{ secrets.XAI_API_KEY }}
+          CLOUDFLARE_API_TOKEN: ${{ secrets.CLOUDFLARE_WORKERS_AI_SYNC_API_TOKEN }}
+          CLOUDFLARE_ACCOUNT_ID: ${{ secrets.CLOUDFLARE_WORKERS_AI_SYNC_ACCOUNT_ID }}
           CLOUDFLARE_WORKERS_AI_SYNC_ACCOUNT_ID: ${{ secrets.CLOUDFLARE_WORKERS_AI_SYNC_ACCOUNT_ID }}
           CLOUDFLARE_WORKERS_AI_SYNC_API_TOKEN: ${{ secrets.CLOUDFLARE_WORKERS_AI_SYNC_API_TOKEN }}
 

README.md

@@ -22,6 +22,17 @@ You can access this data through an API.
 curl https://models.dev/api.json
 ```
 
+Specialized model types are omitted by default. Filter by one or more
+comma-separated model types, or use `all` for the complete catalog:
+
+```bash
+curl "https://models.dev/api.json?type=decision"
+curl "https://models.dev/api.json?type=all"
+```
+
+The currently supported model type is `decision`. The `type` parameter is also
+available on `models.json`, `catalog.json`, and `model-schema.json`.
+
 Use the **Model ID** field to do a lookup on any model; it's the identifier used by [AI SDK](https://ai-sdk.dev/).
 
 Provider-agnostic model metadata is available separately:
@@ -270,6 +281,7 @@ Models must conform to the following schema, as defined in `packages/core/src/sc
 **Model Schema:**
 
 - `name`: String — Display name of the model
+- `type` _(optional)_: String — Specialized model behavior; currently supports `decision`
 - `attachment`: Boolean — Supports file attachments
 - `reasoning`: Boolean — Supports reasoning / chain-of-thought
 - `tool_call`: Boolean - Supports tool calling

labs/vispark/lab.toml

@@ -0,0 +1 @@
+description = "Vispark builds India-centered foundational multimodal intelligence (Vision) with 1M-token context, served via Vispark Lab's API."

models/ai21/jamba-large.toml

@@ -0,0 +1,24 @@
+name = "Jamba Large"
+description = "AI21's hybrid SSM-Transformer long-context model for enterprise agents and grounded generation"
+family = "jamba"
+release_date = "2025-07-01"
+last_updated = "2025-07-01"
+attachment = false
+reasoning = false
+tool_call = true
+structured_output = true
+temperature = true
+knowledge = "2024-08-22"
+open_weights = true
+
+[limit]
+context = 256_000
+output = 4_096
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/ai21labs/AI21-Jamba-Large-1.6"

models/ai21/jamba-mini.toml

@@ -0,0 +1,24 @@
+name = "Jamba Mini"
+description = "AI21's efficient, lightweight hybrid SSM-Transformer model for a wide range of tasks"
+family = "jamba"
+release_date = "2026-01-01"
+last_updated = "2026-01-01"
+attachment = false
+reasoning = false
+tool_call = true
+structured_output = true
+temperature = true
+knowledge = "2024-08-22"
+open_weights = true
+
+[limit]
+context = 256_000
+output = 4_096
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/ai21labs/AI21-Jamba-Mini-1.6"

models/alibaba/qwen3.8-flash-next.toml

@@ -0,0 +1,36 @@
+# Sources (accessed 2026-08-28):
+# https://huggingface.co/Qwen/Qwen3.8-Flash-Next
+# https://huggingface.co/api/models/Qwen/Qwen3.8-Flash-Next
+# https://qwen.ai/blog?id=qwen3.8-flash-next
+# Hub lastModified 2026-08-27T05:03:36Z is the open-weight drop (Do not use
+# Hub createdAt, staged countdown page).
+# Experimental preview of the Qwen4 architecture (Qwen4Exp): hybrid
+# Gated DeltaNet + Qwen Sparse Attention, 512 experts (10 routed + 1 shared),
+# 125B total with 6B active plus 51B n-gram embedding and 4B MTP.
+# Thinking always on: reasoning_effort low|medium|xhigh (default xhigh).
+# Native context 262K, extensible up to 1M tokens.
+
+name = "Qwen3.8 Flash Next"
+description = "Open-weight experimental preview of the Qwen4 architecture: hybrid-attention MoE (125B total, 6B active) with vision encoder for coding, agent tasks, and image and video understanding"
+family = "qwen"
+release_date = "2026-08-27"
+last_updated = "2026-08-27"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = true
+license = "qwen-community-1.0"
+
+[limit]
+context = 262_144
+output = 131_072
+
+[modalities]
+input = ["text", "image", "video"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/Qwen/Qwen3.8-Flash-Next"

models/alibaba/qwen3.8-max-0902.toml

@@ -0,0 +1,28 @@
+# Sources (accessed 2026-09-03):
+# https://www.qwencloud.com/models/qwen3.8-max-0902
+# https://www.alibabacloud.com/help/en/model-studio/qwen3-8-max
+# https://docs.qwencloud.com/developer-guides/text-generation/thinking
+# Snapshot of qwen3.8-max (alias qwen3.8-max-2026-09-02). Same limits,
+# hybrid thinking, modalities, and tool ecosystem as qwen3.8-max; post-trained
+# for stronger coding, collaborative agents, and vision/document understanding.
+# PDF input: inherits Completions PDF理解 / document parsing from the Max line
+# (same as models/alibaba/qwen3.8-max.toml).
+
+name = "Qwen3.8 Max 0902"
+description = "2026-09-02 upgraded snapshot of Qwen3.8 Max with stronger coding, collaborative agents, and multimodal document understanding"
+family = "qwen"
+release_date = "2026-09-02"
+last_updated = "2026-09-02"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+open_weights = false
+
+[limit]
+context = 1_000_000
+output = 131_072
+
+[modalities]
+input = ["text", "image", "video", "pdf"]
+output = ["text"]

models/alibaba/qwen3.8-omni-flash.toml

@@ -0,0 +1,20 @@
+# Sources (accessed 2026-09-18):
+# https://www.alibabacloud.com/help/en/model-studio/qwen3-8-omni-flash
+# https://www.alibabacloud.com/help/en/model-studio/qwen-omni
+name = "Qwen3.8 Omni Flash"
+description = "Qwen omni model for text, vision, audio, and multimodal agent tasks"
+family = "qwen"
+release_date = "2026-09-17"
+last_updated = "2026-09-17"
+attachment = true
+reasoning = true
+tool_call = true
+open_weights = false
+
+[limit]
+context = 1_000_000
+output = 131_072
+
+[modalities]
+input = ["text", "image", "audio", "video"]
+output = ["text"]

models/amazon/nova-2-lite.toml

@@ -0,0 +1,22 @@
+# Sources: https://docs.aws.amazon.com/nova/latest/nova2-userguide/what-is-nova-2.html
+# https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-2-lite.html
+# Launch: https://aws.amazon.com/about-aws/whats-new/2025/12/nova-2-foundation-models-amazon-bedrock/
+name = "Nova 2 Lite"
+description = "Multimodal reasoning model for visual analysis, planning, and tool use"
+family = "nova"
+release_date = "2025-12-02"
+last_updated = "2025-12-01"
+attachment = true
+reasoning = true
+temperature = true
+knowledge = "2025-10"
+tool_call = true
+open_weights = false
+
+[limit]
+context = 1_000_000
+output = 65_536
+
+[modalities]
+input = ["text", "image", "video", "pdf"]
+output = ["text"]

models/amazon/nova-lite.toml

@@ -0,0 +1,24 @@
+# Sources: https://docs.aws.amazon.com/nova/latest/userguide/what-is-nova.html
+# https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-lite.html
+# Launch: https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/
+# PDF document input is supported through Bedrock Converse.
+# Output: the Nova V1 guide's 10K is confirmed by US Bedrock Converse boundary checks (2026-09-09).
+name = "Nova Lite"
+description = "Efficient model for low-latency assistance, extraction, and routine automation"
+family = "nova-lite"
+release_date = "2024-12-03"
+last_updated = "2024-12-03"
+attachment = true
+reasoning = false
+temperature = true
+knowledge = "2024-10"
+tool_call = true
+open_weights = false
+
+[limit]
+context = 300_000
+output = 10_000
+
+[modalities]
+input = ["text", "image", "video", "pdf"]
+output = ["text"]