RAG and retrieval
Results are only compared within a matching group, so a model tested one way is never ranked against a model tested another way.
Retrieve relevant context and ground an answer
Ranking groups use separate scales. Compare configurations only inside the same group; EvalRank never flattens unlike systems into one leaderboard.
Cross-benchmark consensus
One ranking of models across this topic's admitted benchmarks, weighted the same way decisions are. Each score is the model's share of the best result.
| Rank | Model | Share of best | Benchmarks |
|---|---|---|---|
| 1 | GPT-6 Astra | 100.0% | 1 |
| 2 | GPT-6.1 Sol | 97.4% | 1 |
| 3 | Gemma 4 26B A4B | 97.1% | 1 |
| 4 | GPT-6 Luna | 95.7% | 1 |
| 5 | Gemma 4 31B | 94.7% | 1 |
| 6 | Qwen 3.8 Max | 92.1% | 1 |
| 7 | Gemini 4 Argon | 91.6% | 1 |
| 8 | Grok 4.7 | 85.6% | 1 |
| 9 | Gemini 3.8 Flash | 85.1% | 1 |
| 10 | Claude Opus 5.5 | 77.9% | 1 |
| 11 | GPT-5.6 Terra | 75.5% | 1 |
| 12 | Claude Haiku 5.5 | 71.1% | 1 |
| 13 | Gemini 3.1 Pro Preview | 70.7% | 1 |
| 14 | GLM-5.3-Flash | 70.0% | 1 |
| 15 | gpt-oss-120b | 25.1% | 1 |
Model configuration · Direct prompt
Exploratory - not a ranked winner yetrg-rag-retrieval-model-configuration-direct-prompt-model-configuration-v1Explorer evidence
FACTS Grounding v2
Single source. Observed ; fresh until .
| Rank | Configuration | Capability | Uncertainty | Families |
|---|---|---|---|---|
| 1 | GPT-6 Astra facts-grounding-v2:gpt-6-astraconfig_133c9e505fd788462ac768c4805bef5b5fe59e2324bb1f8fff6e845bcddb3872 | 80% | Unknown | 1 |
| 2 | GPT-6.1 Sol facts-grounding-v2:gpt-6.1-solconfig_022373fecf0bf7c463088eb9899c7a3d9eb4bce9cfd24110dee5d64340258fc8 | 78% | Unknown | 1 |
| 3 | Gemma 4 26B A4B facts-grounding-v2:gemma-4-26b-a4b-itconfig_55069a9aaccbd5ee73d7a243cd0c8df03b39b412016cf1a0b7b837edafff174d | 78% | Unknown | 1 |
| 4 | GPT-6 Luna facts-grounding-v2:gpt-6-lunaconfig_dfbe5630ca5e9b5b949f0ac1c2d16d02b8386ac997ef96002c20dcbcbf44251b | 77% | Unknown | 1 |
| 5 | Gemma 4 31B facts-grounding-v2:gemma-4-31b-itconfig_fa8dd1d921924622402184bc1cc37f6f1c4a9ab5a8550e75c01250d3132198cf | 76% | Unknown | 1 |
| 6 | Qwen 3.8 Max facts-grounding-v2:qwen3.8-maxconfig_8fa55caebd27c26e9eb72a531ca084b0b7eb587def26ed79829c31672d534a13 | 74% | Unknown | 1 |
| 7 | Gemini 4 Argon facts-grounding-v2:gemini-4-argonconfig_d111583ea85b8d585e1eff81d6a64ccb192e4f584cb7df57af454e5c2b45e1de | 73% | Unknown | 1 |
| 8 | Grok 4.7 facts-grounding-v2:grok-4.7config_086c38708f99bbe547c2740a59f374dc1c52ea959656105f1b84b8bc40495637 | 69% | Unknown | 1 |
| 9 | Gemini 3.8 Flash facts-grounding-v2:gemini-3.8-flashconfig_289038458a43165e74dc5110c147d0ee5e4582c3c45ee8e794afb02b454cbea6 | 68% | Unknown | 1 |
| 10 | Claude Opus 5.5 facts-grounding-v2:claude-opus-5-5-defaultconfig_d2c7c107ab0570cb61aecf7976c1bf66a5f97bfb7edaa45f27a21886281c4c92 | 62% | Unknown | 1 |
| 11 | GPT-5.6 Terra facts-grounding-v2:gpt-5.6-terraconfig_c000ae6e53b83f053539736253672afb6731d79b61281f862e6b43a61ab41378 | 61% | Unknown | 1 |
| 12 | Claude Haiku 5.5 facts-grounding-v2:claude-haiku-5-5config_1c2522cedfd238e9b5868cd71302f698243450ff10f5dd811765a2283820cbb0 | 57% | Unknown | 1 |
| 13 | Gemini 3.1 Pro Preview facts-grounding-v2:gemini-3.1-pro-previewconfig_5e7d3db54d056747b7f5db367b48e115c38aff0a37a99c447039ea568c28e338 | 57% | Unknown | 1 |
| 14 | GLM-5.3-Flash facts-grounding-v2:glm-5.3-flashconfig_8d840320d281170c66881abda6aacd6a3f818b6431d7a49ff4b96548d6c71112 | 56% | Unknown | 1 |
| 15 | gpt-oss-120b facts-grounding-v2:gpt-oss-120bconfig_a76962bdae4d7535750df3b65733135f2ed5f454f059573b12adbef178ba23ec | 20% | Unknown | 1 |
Not enough independent evidence for a top set
- Calibration has not been validated
- Independent benchmark-family coverage is below the required threshold