Best AI model for coding
The best AI model for coding on current benchmarks is Claude Opus 5.5 (Anthropic): as of October 19, 2026 it has the highest FrontierCode v1.1 score TensorFeed tracks, 54.6%, 1.1 points ahead of Claude Fable 5 (53.5%), among 30 models scored. The top four sit within 1.3 points of each other. It also leads Terminal-Bench 4.0 (65.2%).
Ranked on FrontierCode v1.1. Benchmark data as of October 19, 2026; prices as of October 19, 2026.
FrontierCode v1.1: 52.1%. $2.00 / $10.00 per 1M tokens.
Ranking on FrontierCode v1.1
Each score links to its source; vendor-reported and independent figures are labeled and are not interchangeable.
| # | Model | Provider | FrontierCode v1.1 | Price in / out | TFII | Terminal-Bench 4.0 | SWE-bench |
|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 | Anthropic | 54.6%independent | $4.00 / $20.00 | 97.1 | 65.15% | - |
| 2 | Claude Fable 5 | Anthropic | 53.5% | $10.00 / $50.00 | 81.6 | 41.41% | 95% |
| 3 | Claude Opus 5 | Anthropic | 53.4%vendor | $5.00 / $25.00 | 83.9 | 53.54% | 97% |
| 4 | GPT-6 Astra | OpenAI | 53.3%vendor | $10.00 / $50.00 | 83.3 | 59.6% | - |
| 5 | Claude Sonnet 5.5 | Anthropic | 52.1%independent | $2.00 / $10.00 | 93.7 | 64.14% | - |
| 6 | Claude Fable 5.1 | Anthropic | 50.9%independent | $10.00 / $50.00 | 89.9 | 58.08% | - |
| 7 | GPT-6.1 Sol | OpenAI | 50.2%independent | $2.00 / $10.00 | - | 55.05% | - |
| 8 | GPT-6 Sol | OpenAI | 49.3%independent | $2.00 / $10.00 | - | 44.44% | - |
| 9 | Grok 4.6 | xAI | 48%independent | $2.00 / $6.00 | 83.2 2024 set | 17.17% | 95.6% |
| 10 | GPT-5.6 Sol | OpenAI | 47.5%independent | $4.00 / $20.00 | 74.5 | 37.88% | 96.2% |
| 11 | Claude Opus 4.8 | Anthropic | 46.5% | $5.00 / $25.00 | 67.9 | 23.23% | 88.6% |
| 12 | Kimi K3 | Moonshot AI | 44.2%independent | $3.00 / $15.00 | 65.1 | 17.17% | 93.4% |
| 13 | Gemini 3.7 Flash | 43.6%independent | - | 52.3 | 12.12% | 80.8% | |
| 14 | GPT-5.5 | OpenAI | 43%independent | $5.00 / $30.00 | 69.8 2024 set | - | 82.6% |
| 15 | Claude Sonnet 5 | Anthropic | 42.7%independent | $2.00 / $10.00 | 59.2 | 10.61% | 85.2% |
| 16 | GPT-6 Luna | OpenAI | 42.4%independent | $0.10 / $0.50 | - | 13.64% | - |
| 17 | Grok 4.5 | xAI | 42.4%independent | $2.00 / $6.00 | 80.0 2024 set | 8.59% | 86.6% |
| 18 | GPT-5.6 Terra | OpenAI | 41.3%independent | $2.00 / $12.00 | 60.9 | 22.73% | 95.4% |
| 19 | Gemini 3.8 Flash | 41.2%independent | $0.75 / $3.75 | 79.0 2024 set | 19.7% | 80% | |
| 20 | GLM-5.3 | Z.ai | 40.1%independent | $1.40 / $4.40 | 71.8 | 38.89% | 95.4% |
| 21 | GPT-5.6 Luna | OpenAI | 39.8%independent | $0.20 / $1.20 | 53.2 | 11.62% | 93% |
| 22 | Claude Opus 4.7 | Anthropic | 38.5%independent | $5.00 / $25.00 | 69.8 2024 set | - | 87.6% |
| 23 | Gemini 3.6 Flash | 34.4%independent | $0.75 / $3.75 | 78.1 2024 set | - | 79.6% | |
| 24 | GLM-5.3 Flash | Z.ai | 31.8%independent | $0.15 / $0.50 | 62.8 | 37.37% | 92% |
| 25 | Claude Opus 4.6 | Anthropic | 26.6%independent | $5.00 / $25.00 | 67.8 2024 set | - | 80.8% |
| 26 | GLM-5.2 | Z.ai | 24.5%independent | $1.40 / $4.40 | 75.6 2024 set | - | 82.8% |
| 27 | Claude Sonnet 4.6 | Anthropic | 24.3%independent | $3.00 / $15.00 | 64.6 2024 set | - | 79.6% |
| 28 | MiniMax M3 | MiniMax | 14.7%independent | $0.30 / $1.20 | 29.8 | 1.01% | 80.5% |
| 29 | Inkling | Thinking Machines | 14%independent | - | 35.7 | 0.51% | 77.6% |
| 30 | Mistral Medium 3.5 | Mistral | 8%independent | $1.50 / $7.50 | - | - | 77.6% |
Prices are list prices per 1M input / output tokens. TFII is the headline TensorFeed Intelligence Index score; scores marked 2024 set come from older benchmarks and are not comparable with the rest.
How this is ranked
FrontierCode v1.1 asks whether a maintainer would merge the patch, not only whether it passes tests: open-source maintainers wrote the tasks against their own repositories and grade scope, regression safety, test quality and project conventions. It still separates the 2026 flagships. SWE-bench Verified, the old default, is saturated (Vals AI archived its board on September 1, 2026, and the newest flagships do not report it), so it is shown for reference only. Terminal-Bench 4.0 sits alongside as the broader test of long, unattended work in a terminal.
FrontierCode scores move with reasoning effort, and the board publishes a figure per effort level, so a gap of a point or two between leaders is not decisive. Check the effort level in the provenance note before you compare.
The full board, scoring notes and score ranges are on the FrontierCode v1.1 leaderboard. The same scores are free as JSON at /api/benchmarks.
Questions
What is the best AI model for coding?
The best AI model for coding on current benchmarks is Claude Opus 5.5 (Anthropic): as of October 19, 2026 it has the highest FrontierCode v1.1 score TensorFeed tracks, 54.6%, 1.1 points ahead of Claude Fable 5 (53.5%), among 30 models scored. The top four sit within 1.3 points of each other. It also leads Terminal-Bench 4.0 (65.2%).
Which of the top AI models for coding is cheapest?
Of the five highest FrontierCode v1.1 scores TensorFeed tracks, the lowest-priced model with a list price is Claude Sonnet 5.5 (52.1%), at $2.00 input and $10.00 output per 1M tokens as of October 19, 2026.
Why rank AI models for coding on FrontierCode v1.1?
FrontierCode v1.1 asks whether a maintainer would merge the patch, not only whether it passes tests: open-source maintainers wrote the tasks against their own repositories and grade scope, regression safety, test quality and project conventions. It still separates the 2026 flagships. SWE-bench Verified, the old default, is saturated (Vals AI archived its board on September 1, 2026, and the newest flagships do not report it), so it is shown for reference only. Terminal-Bench 4.0 sits alongside as the broader test of long, unattended work in a terminal.