The Inference Report

August 9, 2026

Daily rankings from SWE-rebench, a benchmark designed to fairly compare LLM capabilities on real-world software engineering tasks. Unlike other evaluations, it uses a standardized scaffolding for all models, continuously updates its dataset to prevent contamination, and runs each model five times to account for stochastic variance.

#ModelScore
1AnthropicFable 5 [high]Model64.5%± 1.41%
2GrokGrok 4.5 [high]Model63.8%± 0.60%
3AnthropicOpus 5 [high]Model63.4%± 1.35%
4Z.aiGLM-5.2 [high]Model62.9%± 1.19%
5OpenAIGPT-5.6 Sol [medium]Model62.3%± 1.83%
6JunieJunieAgent61.8%± 0.54%
7AnthropicClaude CodeAgent60.4%± 1.03%
8OpenAICodexAgent58.0%± 1.29%
9AnthropicSonnet 5 [high]Model56.8%± 0.94%
10CursorCursorAgent51.7%± 0.84%

Artificial Analysis composite index across coding, math, and reasoning benchmarks.

#ModelScoretok/s$/1M
1Claude Opus 563.148$10.00
2Claude Fable 562.158$20.00
3GPT-5.6 Sol60.967$11.25
4Kimi K359.737$6.00
5Qwen3.8 Max58.180$3.00
6Claude Opus 4.857.30$10.00
7Muse Spark 1.256.80$2.00
8GPT-5.6 Terra56.6116$4.50
9GPT-5.556.30$11.25
10Grok 4.555.848$3.00

Output tokens per second — higher is faster. Minimum intelligence score of 40.

#Modeltok/s
1Gemini 3.5 Flash226
2Qwen3.7 Max197
3Muse Spark 1.1192
4Gemini 3.6 Flash184
5GPT-5.6 Luna161
6Inkling Small144
7Gemini 3.1 Pro Preview126
8Nex-N2-Pro124
9GPT-5.6 Terra116
10GPT-5.3 Codex112

Blended cost per 1M tokens (3:1 input/output) — lower is cheaper. Minimum intelligence score of 40.

#Model$/1M
1DeepSeek V4 Flash$0.168
2DeepSeek V4 Flash 0731$0.175
3Hy3$0.241
4GPT-5.6 Luna$0.45
5MiniMax-M3$0.525
6Inkling Small$0.525
7DeepSeek V4 Pro$0.544
8MiMo-V2.5-Pro$0.544
9Nex-N2-Pro$1.00
10Qwen3.6 Plus$1.13