Compare published benchmark performance with estimated workload cost and visible evidence coverage.
Interactive Tool
| Model | Performance rank | Performance score | Coverage | Workload cost | Cost / score point |
|---|---|---|---|---|---|
| gpt-oss-120bOpenAI | 33 | 21.28 | 27.27% | $5.68 | $0.27 |
| GLM 5.3 FlashZ.ai | 18 | 34.28 | 36.36% | $12.50 | $0.36 |
| GPT-6 LunaOpenAI | 8 | 46.78 | 54.55% | $20.00 | $0.43 |
| GPT-4o MiniOpenAI | 34 | 20.75 | 36.36% | $13.50 | $0.65 |
| Llama 4 Scout 17B 16E InstructMeta | 28 | 23.86 | 36.36% | $16.00 | $0.67 |
| DeepSeek V4.1 FlashDeepSeek | 19 | 32.98 | 36.36% | $27.00 | $0.82 |
| Gemma 4 31BGoogle DeepMind | 24 | 26.21 | 27.27% | $28.00 | $1.07 |
| MiniMax M3MiniMax | 12 | 41.32 | 45.45% | $50.00 | $1.21 |
| Llama 4 Maverick 17B 128E InstructMeta | 26 | 25.55 | 36.36% | $31.80 | $1.24 |
| GPT-5.4 NanoOpenAI | 21 | 29.84 | 45.45% | $45.00 | $1.51 |
| Mistral Large 3Mistral AI | 30 | 23.71 | 27.27% | $40.00 | $1.69 |
| Qwen3.8 27BQwen | 16 | 37.90 | 45.45% | $64.00 | $1.69 |
| Gemini 3.5 Flash LiteGoogle DeepMind | 17 | 37.29 | 54.55% | $80.00 | $2.15 |
| Qwen3.7 PlusQwen | 23 | 26.42 | 27.27% | $57.60 | $2.18 |
| Inkling SmallThinking Machines Lab | 22 | 28.77 | 36.36% | $69.00 | $2.40 |
| Gemini 3.8 FlashGoogle DeepMind | 7 | 50.85 | 54.55% | $150.00 | $2.95 |
| DeepSeek V4 ProDeepSeek | 27 | 24.67 | 27.27% | $96.88 | $3.93 |
| InklingThinking Machines Lab | 11 | 41.78 | 54.55% | $176.00 | $4.21 |
| Grok 4.20 Multi AgentxAI | 13 | 40.75 | 45.45% | $175.00 | $4.29 |
| GPT-5.4 MiniOpenAI | 20 | 32.20 | 45.45% | $165.00 | $5.12 |
| Qwen3.8 MaxQwen | 10 | 43.94 | 45.45% | $264.02 | $6.01 |
| GPT-6.1 SolOpenAI | 4 | 62.49 | 63.64% | $400.00 | $6.40 |
| GLM 5.3Z.ai | 25 | 26.02 | 27.27% | $170.00 | $6.53 |
| Kimi K3Moonshot AI | 14 | 39.81 | 45.45% | $262.62 | $6.60 |
| Gemini 3.1 ProGoogle DeepMind | 3 | 66.47 | 72.73% | $440.00 | $6.62 |
| GPT-5.6 TerraOpenAI | 5 | 60.65 | 63.64% | $440.00 | $7.25 |
| Claude Sonnet 5Anthropic | 6 | 51.53 | 54.55% | $400.00 | $7.76 |
| Grok 4.7xAI | 15 | 39.45 | 45.45% | $320.00 | $8.11 |
| Claude Haiku 4.5Anthropic | 32 | 21.88 | 36.36% | $200.00 | $9.14 |
| GPT-4oOpenAI | 29 | 23.76 | 36.36% | $225.00 | $9.47 |
| Claude Opus 5.5Anthropic | 9 | 44.85 | 45.45% | $800.00 | $17.84 |
| Command ACohere | 31 | 22.42 | 36.36% | $450.00 | $20.07 |
| Claude Fable 5.1Anthropic | 1 | 71.90 | 72.73% | $2,000.00 | $27.82 |
| GPT-6 AstraOpenAI | 2 | 71.34 | 72.73% | $2,000.00 | $28.03 |
Compare rows with similar benchmark coverage. A low ratio is a screening signal, not a workload-specific quality guarantee.
Inputs
| Input | How it is used |
|---|---|
| Monthly workload | Input and output tokens used to estimate each model's cost. |
| Minimum coverage | Required share of the published ranking benchmark set. |
The table joins source-linked token prices to Model Markets' verified benchmark ranking. Missing prices or benchmark rows stay unknown and do not receive fabricated ratios.
Compare rows within similar evidence coverage. A cheap model with sparse benchmark participation can look efficient simply because the hard tests are missing.
Boundaries
Catalog values retain their source and freshness on the linked model, provider, benchmark, or comparison page. Editable scenario assumptions are not Model Markets measurements.
Continue the Analysis
| Tool | Next question |
|---|---|
| Benchmark Comparison | Inspect like-for-like published results without blending incompatible metrics or versions. |
| Model Selector | Turn workload constraints into a short, inspectable model shortlist instead of a universal best-model claim. |
| LLM Cost Calculator | Price one explicit token workload against an exact published provider endpoint. |
Questions
Screen for models that combine useful published performance with acceptable token economics. It returns a sortable frontier-style table with benchmark score, coverage, workload cost, and cost per performance point.
Where the calculation needs model facts, it uses the current Model Markets catalog snapshot updated 2026-10-19. User-entered assumptions remain clearly editable, and unsupported values stay unknown rather than being inferred.
Benchmark aggregation is not a workload-specific quality guarantee. Lowest token prices can come from different providers. Latency and reliability are not part of the ratio. Open the linked canonical records and primary sources before making a production or purchasing decision.
