The Inference Report

March 29, 2026

The Value Frontier

Price vs Intelligence — lower price, higher score is better.

Fast and Smart

Speed vs Intelligence — higher on both axes is better.

Efficiency

Speed vs Price — faster and cheaper is better.

SWE-rebench

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.

#	Model	Score
1	Claude Opus 4.6	65.3%
2	gpt-5.2-2025-12-11-medium	64.4%
3	GLM-5	62.8%
4	gpt-5.4-2026-03-05-medium	62.8%
5	Gemini 3.1 Pro Preview	62.3%
6	DeepSeek-V3.2	60.9%
7	Claude Sonnet 4.6	60.7%
8	Claude Sonnet 4.5	60.0%
9	Qwen3.5-397B-A17B	59.9%
10	Step-3.5-Flash	59.6%

Intelligence Index — Leaderboard

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

#	Model	Score	tok/s	$/1M
1	GPT-5.4	57.2	84	$5.63
2	Gemini 3.1 Pro Preview	57.2	115	$4.50
3	GPT-5.3 Codex	54	81	$4.81
4	Claude Opus 4.6	53	56	$10.00
5	Claude Sonnet 4.6	51.7	66	$6.00
6	GPT-5.2	51.3	66	$4.81
7	GLM-5	49.8	60	$1.55
8	Claude Opus 4.5	49.7	64	$10.00
9	MiniMax-M2.7	49.6	44	$0.525
10	MiMo-V2-Pro	49.2	83	$1.50

Speed — Leaderboard

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

#	Model	tok/s
1	Grok 4.20 Beta 0309	238
2	GPT-5.4 mini	219
3	GPT-5 Codex	216
4	Gemini 3 Flash Preview	180
5	GPT-5.4 nano	178
6	GPT-5.1 Codex	139
7	Qwen3.5 122B A10B	135
8	MiMo-V2-Flash	134
9	GPT-5.2 Codex	123
10	Gemini 3 Pro Preview	117

Price — Leaderboard

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

#	Model	$/1M
1	MiMo-V2-Flash	$0.15
2	DeepSeek V3.2	$0.315
3	GPT-5.4 nano	$0.463
4	MiniMax-M2.7	$0.525
5	MiniMax-M2.5	$0.525
6	GPT-5 mini	$0.688
7	Qwen3.5 27B	$0.825
8	GLM-4.7	$1.00
9	Kimi K2 Thinking	$1.07
10	Qwen3.5 122B A10B	$1.10