InferenceBench

Independent AI Providers Benchmark.
Same model. Different infrastructure. Different result.

[InferenceBench @InferenceBench]

Kimi K3 via OpenRouter raised its input price 76 percent, 1.70 to 3.00 per million tokens. Evidence: https://t.co/9aYgnPMNny

[InferenceBench @InferenceBench]

Reliability for GLM-4.7 via OpenRouter fell 3.2 points to 96.8 percent in our benchmarks. Evidence: https://t.co/14exWgAyyl

[InferenceBench @InferenceBench]

DeepSeek V4 Pro via OpenRouter raised its input price 125 percent, 0.42 to 0.95 per million tokens. Evidence: https://t.co/1DkHQzeg7R

[InferenceBench @InferenceBench]

Reliability for Mistral Large 3 via OpenRouter fell 61.3 points to 0 percent in our benchmarks. Evidence: https://t.co/yWm5DJ5zHI

[InferenceBench @InferenceBench]

Reliability for GPT-OSS 120B on Together AI fell 9.7 points to 90.3 percent in our benchmarks. Evidence: https://t.co/vW8HqsCijG

[InferenceBench @InferenceBench]

Reliability for Mistral Large 3 via OpenRouter fell 100 points to 0 percent in our benchmarks. Evidence: https://t.co/yWm5DJ5zHI

[InferenceBench @InferenceBench]

Reliability for GLM-5.2 via OpenRouter fell 3.2 points to 96.8 percent in our benchmarks. Evidence: https://t.co/kOIbrOB6L6

[InferenceBench @InferenceBench]

Reliability for Qwen3-235B via Cortecs fell 3.2 points to 96.8 percent in our benchmarks. Evidence: https://t.co/Ut9X4Nfq9C

[InferenceBench @InferenceBench]

DeepSeek V4 Flash via Cortecs cut its input price 40 percent, 0.08 to 0.05 per million tokens. Evidence: https://t.co/URZnjoKlbv

[InferenceBench @InferenceBench]

Llama 3.3 70B via Cortecs raised its output price 82 percent, 0.36 to 0.65 per million tokens. Evidence: https://t.co/BDGMSw2G0E

[InferenceBench @InferenceBench]

Llama 3.3 70B via Cortecs raised its input price 460 percent, 0.12 to 0.65 per million tokens. Evidence: https://t.co/BDGMSw2G0E

[InferenceBench @InferenceBench]

GLM-5.2 via Cortecs cut its output price 29 percent, 3.77 to 2.68 per million tokens. Evidence: https://t.co/kFUVQukTIm

[InferenceBench @InferenceBench]

Reliability for GPT-OSS 120B on Together AI fell 6.5 points to 87.1 percent in our benchmarks. Evidence: https://t.co/vW8HqsCijG

[InferenceBench @InferenceBench]

Reliability for GLM-5.3 via OpenRouter fell 3.2 points to 96.8 percent in our benchmarks. Evidence: https://t.co/WETwTf5lNJ

[InferenceBench @InferenceBench]

Reliability for Kimi K3 on Together AI fell 6.5 points to 93.5 percent in our benchmarks. Evidence: https://t.co/3LULwGT2lz

[InferenceBench @InferenceBench]

Reliability for GPT-OSS 120B on Together AI fell 29.0 points to 71.0 percent in our benchmarks. Evidence: https://t.co/vW8HqsCijG

[InferenceBench @InferenceBench]

DeepSeek V4 Flash via OpenRouter raised its input price 37 percent, 0.07 to 0.09 per million tokens. Evidence: https://t.co/BeL2rbEH7p

[InferenceBench @InferenceBench]

Reliability for Llama 3.3 70B via Cortecs fell 9.7 points to 90.3 percent in our benchmarks. Evidence: https://t.co/BDGMSw2G0E

[InferenceBench @InferenceBench]

GLM-5.3 via OpenRouter cut its output price 22 percent, 4.40 to 3.43 per million tokens. Evidence: https://t.co/WETwTf5lNJ

[InferenceBench @InferenceBench]

GLM-5.3 via OpenRouter cut its input price 22 percent, 1.40 to 1.09 per million tokens. Evidence: https://t.co/WETwTf5lNJ