feat(models): expand benchmark coverage

[adamdotdevin]

Popular models on opencode.ai/data, including Muse Spark 1.2/1.3, had no benchmark metadata. Add 402 sourced benchmark results across 26 models from 12 labs, filling 21 previously empty entries and expanding five others.

Prioritized popular models, then recent releases with first-party evaluation tables. Preserve benchmark versions, harnesses, effort levels, tool use, scoring rules, dataset corrections, and checkpoint distinctions where published. Every result includes its source; existing results and other model/provider metadata are unchanged. Hosted aliases without an exact published evaluation (such as Qwen3.8 Flash versus Flash-Next) and unidentified models remain unfilled.

Validation: bun validate, SDK generation/typecheck and all 25 SDK tests pass. A generated-catalog audit confirms all 402 additions survive parsing and existing metadata is unchanged. Core schema/generation tests: 19 pass; the missing-weights-links test fails identically on unchanged dev (41 existing models). Benchmark sourcing and normalization tests pass.

Coverage and primary sources

Model	Added results	Sources
alibaba/qwen3.8-27b	29	source 1
alibaba/qwen3.8-flash-next	27	source 1
alibaba/qwen3.8-max	32	source 1
anthropic/claude-fable-5-1	9	source 1
anthropic/claude-opus-5-5	9	source 1
deepseek/deepseek-v4-flash	21	source 1
deepseek/deepseek-v4-flash-vision-exp	11	source 1
deepseek/deepseek-v4-pro	21	source 1
deepseek/deepseek-v4-pro-0813	11	source 1
deepseek/deepseek-v4.1-flash	34	source 1
google/gemini-3.8-flash	15	source 1
inclusionai/ling-3.0-flash-fin	9	source 1
meta/muse-spark-1.2	14	source 1, source 2
meta/muse-spark-1.3	12	source 1
nvidia/nemotron-3.5-lightning	14	source 1
openai/gpt-6-luna	1	source 1
openai/gpt-6-sol	4	source 1
tencent/hy3	30	source 1
xai/grok-4.6	8	source 1
xai/grok-4.7	8	source 1
xiaomi/mimo-v2.5	12	source 1
xiaomi/mimo-v2.5-pro	15	source 1
xiaomi/mimo-v2.6-flash	16	source 1
xiaomi/mimo-v2.6-pro	17	source 1
zhipuai/glm-5.3	17	source 1
zhipuai/glm-5.3-flash	6	source 1

[github-actions]

Action items

• [high] [possible mistake] models/xiaomi/mimo-v2.5-pro.toml:50 - Check: Benchmark source must evaluate the filed model or an explicitly labeled comparison column. Why: All 15 added results use source = "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL" while the file and its [[weights]] URL are MiMo-V2.5-Pro, with no leading comment explaining the cross-checkpoint attribution. Action: Verify the Flash-RL page actually publishes a V2.5-Pro column; if so add a leading top-of-file comment stating the Pro-2.5 column plus column-identifying variant/dataset, otherwise replace with the V2.5-Pro source or remove.
• [high] [possible mistake] models/xiaomi/mimo-v2.6-pro.toml:29 - Check: Benchmark source must evaluate the filed checkpoint or an explicitly labeled comparison column. Why: All 17 added results use .../MiMo-V2.6-Flash-RL while [[weights]] points to .../MiMo-V2.6-Pro-RL, and the leading comment only says "official MiMo-V2.6 RL checkpoint evaluation table" without stating Flash-vs-Pro column. Action: Verify the Flash-RL table actually publishes the Pro column; if so clarify the Pro column in the leading comment and variant, otherwise source from the Pro-RL page or remove.
• [high] [possible mistake] models/meta/muse-spark-1.2.toml:32 - Check: Benchmark source must evaluate the filed model or an explicitly labeled comparison column. Why: 12 of 14 added results use source = "https://research.meta.ai/blog/introducing-muse-spark-1-3" for a muse-spark-1.2 file, with no leading comment stating these are the 1.2 column of the 1.3 scorecard. Action: Verify the 1.3 blog/scorecard actually publishes 1.2 scores and, if so, add a leading comment stating the 1.2 column; otherwise replace with the 1.2 source or remove.
• [high] [possible mistake] models/xai/grok-4.6.toml:28 - Check: Benchmark source must evaluate the filed model or an explicitly labeled comparison column. Why: All 8 added results use source = "https://x.ai/news/grok-4-7" for a grok-4.6 file with no leading comment explaining the cross-model attribution. Action: Verify the 4.7 announcement actually publishes a 4.6 comparison column and, if so, add a leading top-of-file comment stating the 4.6 column; otherwise replace with the 4.6 source or remove.

[github-actions]

No actionable findings.