AI model
performance
report
A concise view of coding completion, cost efficiency, model popularity, and inference provider reliability across the Kilo ecosystem.
- Published
- October 14, 2026
- Benchmark
- Terminal Bench 2.0
- Inference window
- Latest weekly snapshot
KiloBench leaders
Models are ranked by completion rate on Terminal Bench 2.0. Cost per attempt is the observed average for a complete benchmark attempt, including unsuccessful attempts.
| # | Model | Provider | Completion rate | Cost / attempt |
|---|---|---|---|---|
| 1 | GPT-6 Astra ($$$$) 5 attempts | openai | 79.3% | $107.29 |
| 2 | GPT-5.6 Sol 5 attempts | openai | 76.2% | $87.41 |
| 3 | Claude Fable 5.1 ($$$$) 5 attempts | anthropic | 76.2% | $91.41 |
| 4 | DeepSeek V4.1 Flash 5 attempts | deepseek | 75.3% | $2.58 |
| 5 | Gemini 3.8 Flash (50% off) 5 attempts | 75.3% | $112.27 | |
| 6 | GPT-5.5 5 attempts | openai | 74.2% | $72.63 |
| 7 | Grok 4.6 5 attempts | x-ai | 73.0% | $33.83 |
| 8 | Kimi K3 5 attempts | moonshotai | 72.8% | $48.38 |
| 9 | Claude Opus 5 5 attempts | anthropic | 71.5% | $113.54 |
| 10 | Claude Fable 5 ($$$$) 5 attempts | anthropic | 71.0% | $87.52 |
Inference market view
The 20 most-used models are ranked by token volume from the latest Kilo Gateway weekly snapshot. Primary-provider metrics describe the provider carrying the largest traffic share for each model.
| # | Model | Primary provider | Usage volume | Cost / request | Success rate |
|---|---|---|---|---|---|
| 1 | step-5-preview-free 1 provider | StepFun | 1.7T | $0 | 98.56% |
| 2 | nemotron-3-ultra-550b-a55b:free 1 provider | NVIDIA | 1.4T | $0 | 91.69% |
| 3 | dots-3-note-preview:free 1 provider | Atlas Cloud | 599.5B | $0 | 86.02% |
| 4 | laguna-s-2.1:free 1 provider | Poolside | 561.4B | $0 | 95.26% |
| 5 | deepseek-v4.1-flash 7 providers | Novita AI | 202.0B | $0.0031 | 99.49% |
| 6 | nemotron-3-super-120b-a12b:free 1 provider | NVIDIA | 201.7B | $0 | 90.84% |
| 7 | ling-3.1-flash 1 provider | Novita AI | 195.7B | $0.00000 | 91.45% |
| 8 | step-3.7-flash:free 1 provider | StepFun | 163.4B | $0 | 95.54% |
| 9 | glyph-cluster 1 provider | Stealth | 147.6B | $0 | 96.66% |
| 10 | openrouter/free 7 providers | NVIDIA | 94.5B | $0 | 86.66% |
Inference market view / continued
The 20 most-used models are ranked by token volume from the latest Kilo Gateway weekly snapshot. Primary-provider metrics describe the provider carrying the largest traffic share for each model.
| # | Model | Primary provider | Usage volume | Cost / request | Success rate |
|---|---|---|---|---|---|
| 11 | nemotron-3.5-lightning:free 1 provider | NVIDIA | 70.4B | $0 | 74.69% |
| 12 | glm-5.3-flash 6 providers | Friendli AI | 59.1B | $0.0060 | 99.76% |
| 13 | north-mini-code:free 1 provider | Cohere | 54.7B | $0 | 82.94% |
| 14 | gpt-6-luna 3 providers | OpenAI | 42.9B | $0.0033 | 99.93% |
| 15 | ling-3.0-flash-sante:free 1 provider | Novita AI | 42.6B | $0 | 92.79% |
| 16 | mimo-v2.6-pro 3 providers | Xiaomi | 38.1B | $0.0038 | 99.61% |
| 17 | claude-opus-5.5 3 providers | 35.4B | $0.12 | 95.20% | |
| 18 | muse-spark-1.3-contributor 1 provider | Meta | 34.4B | $0.0018 | 99.84% |
| 19 | gpt-6.1-sol 4 providers | OpenAI | 34.0B | $0.0530 | 99.94% |
| 20 | claude-sonnet-5.5 3 providers | 15.6B | $0.0843 | 98.75% |
Cost and reliability
Provider-level traffic share, average request cost, and success rate for the ten most-used models. Traffic share uses the percentage published by Kilo Gateway.
step-5-preview-free
| Provider | Traffic share | Cost / request | Success rate |
|---|---|---|---|
| StepFun | 100.00% | $0 | 98.56% |
nemotron-3-ultra-550b-a55b:free
| Provider | Traffic share | Cost / request | Success rate |
|---|---|---|---|
| NVIDIA | 100.00% | $0 | 91.69% |
dots-3-note-preview:free
| Provider | Traffic share | Cost / request | Success rate |
|---|---|---|---|
| Atlas Cloud | 100.00% | $0 | 86.02% |
laguna-s-2.1:free
| Provider | Traffic share | Cost / request | Success rate |
|---|---|---|---|
| Poolside | 100.00% | $0 | 95.26% |
Cost and reliability
Provider-level traffic share, average request cost, and success rate for the ten most-used models. Traffic share uses the percentage published by Kilo Gateway.
deepseek-v4.1-flash
| Provider | Traffic share | Cost / request | Success rate |
|---|---|---|---|
| Novita AI | 93.93% | $0.0031 | 99.49% |
| Alibaba | 4.81% | $0.0187 | 99.88% |
| Fireworks AI | 1.01% | $0.0049 | 99.87% |
| Baseten | 0.17% | $0.0094 | 99.90% |
| Together AI | 0.06% | $0.0028 | 100.00% |
| Relace | 0.01% | $0.0016 | 100.00% |
| DeepSeek | <0.01% | $0.0020 | 91.76% |
nemotron-3-super-120b-a12b:free
| Provider | Traffic share | Cost / request | Success rate |
|---|---|---|---|
| NVIDIA | 100.00% | $0 | 90.84% |
ling-3.1-flash
| Provider | Traffic share | Cost / request | Success rate |
|---|---|---|---|
| Novita AI | 100.00% | $0.00000 | 91.45% |
step-3.7-flash:free
| Provider | Traffic share | Cost / request | Success rate |
|---|---|---|---|
| StepFun | 100.00% | $0 | 95.54% |
Cost and reliability
Provider-level traffic share, average request cost, and success rate for the ten most-used models. Traffic share uses the percentage published by Kilo Gateway.
glyph-cluster
| Provider | Traffic share | Cost / request | Success rate |
|---|---|---|---|
| Stealth | 100.00% | $0 | 96.66% |
openrouter/free
| Provider | Traffic share | Cost / request | Success rate |
|---|---|---|---|
| NVIDIA | 45.35% | $0 | 86.66% |
| Novita AI | 17.89% | $0 | 83.12% |
| Atlas Cloud | 14.83% | $0 | 82.88% |
| Poolside | 13.26% | $0 | 93.35% |
| Cohere | 7.30% | $0 | 80.87% |
| Liquid | 1.12% | $0 | 59.33% |
| Thinking Machines | 0.24% | $0 | 86.36% |