Skip to main content

Best Value LLM for Knowledge in 2026 — Cost-Adjusted Rankings

Knowledge benchmarks test factual accuracy, expert-level science, and professional comprehension. This ranking divides each model's weighted knowledge score (GPQA, SuperGPQA, MMLU-Pro, HLE, FrontierScience, SimpleQA) by output token price. For RAG pipelines, research assistants, and Q&A systems where accurate knowledge retrieval matters but API costs add up, the value leaders here offer the best accuracy per dollar.

Unless noted otherwise, ranking surfaces on this page use BenchLM's provisional leaderboard lane rather than the stricter sourced-only verified leaderboard.

Bottom line: Knowledge tasks don't require expensive reasoning models. Gemini 3.1 Flash-Lite leads value, and DeepSeek Coder 2.0 offers strong raw knowledge at low cost.

According to BenchLM.ai, Gemini 3.1 Flash-Lite leads this ranking with a score of 98.75, followed by DeepSeek Coder 2.0 (53.15) and Gemini 2.5 Flash (51.27). There is a significant gap between the leading models and the rest of the field.

The best open-weight option is DeepSeek Coder 2.0 (ranked #2 with a score of 53.15). Open-weight models are highly competitive in this category — self-hosting is a viable alternative to proprietary APIs.

This ranking is based on provisional weighted averages across the scoring benchmarks in knowledge tracked by BenchLM.ai. For detailed model profiles, click any model name below. To compare two specific models head-to-head, use the "vs #" links.

What changed

Gemini 3.1 Flash-Lite leads knowledge value — most knowledge capability per dollar.

DeepSeek Coder 2.0 strong raw knowledge scores at very low cost.

Gemini 2.5 Flash good knowledge value with broader model capabilities.

How to choose

Full Rankings (24 models)

Gemini 3.1 Flash-Lite
Google·Proprietary·1M

98.75

Score/$

Score: 39.5 · $0.4/1M

DeepSeek Coder 2.0
DeepSeek·Open Weight·128K

53.15

Score/$

Score: 58.5 · $1.1/1M

Gemini 2.5 Flash
Google·Proprietary·1M

51.27

Score/$

Score: 30.8 · $0.6/1M

4
DeepSeek V3
DeepSeek·Open Weight·128K

42.96

Score/$

Score: 47.3 · $1.1/1M

5
Kimi K2.5
Moonshot AI·Open Weight·128K

25.22

Score/$

Score: 70.6 · $2.8/1M

6
GLM-5.1
Z.AI·Open Weight·203K

19.36

Score/$

Score: 85.2 · $4.4/1M

7
Gemini 3.1 Pro
Google·Proprietary·1M

19.11

Score/$

Score: 95.6 · $5/1M

8
DeepSeek-R1
DeepSeek·Open Weight·128K

18.38

Score/$

Score: 40.3 · $2.19/1M

9
Gemini 3 Flash
Google·Proprietary·1M

17.78

Score/$

Score: 53.3 · $3/1M

10
GPT-5.1
OpenAI·Proprietary·200K

13.93

Score/$

Score: 83.6 · $6/1M

11
Gemini 2.5 Pro
Google·Proprietary·1M

13.55

Score/$

Score: 67.7 · $5/1M

12
Claude Haiku 4.5
Anthropic·Proprietary·200K

12.89

Score/$

Score: 51.5 · $4/1M

13
GPT-5.2
OpenAI·Proprietary·400K

11.66

Score/$

Score: 93.3 · $8/1M

14
GPT-5.1-Codex-Max
OpenAI·Proprietary·400K

10.08

Score/$

Score: 80.7 · $8/1M

15
GPT-5.2-Codex
OpenAI·Proprietary·400K

10.04

Score/$

Score: 80.3 · $8/1M

16
GPT-5.3 Codex
OpenAI·Proprietary·400K

9.44

Score/$

Score: 94.4 · $10/1M

17
Mistral Large 3
Mistral·Proprietary·128K

7.43

Score/$

Score: 44.6 · $6/1M

18
GPT-5.4
OpenAI·Proprietary·1.05M

6.51

Score/$

Score: 97.6 · $15/1M

19
Grok 4.1
xAI·Proprietary·1M

6.31

Score/$

Score: 94.7 · $15/1M

20
Claude Sonnet 4.6
Anthropic·Proprietary·200K

5.65

Score/$

Score: 84.7 · $15/1M

21
Claude Sonnet 4.5
Anthropic·Proprietary·200K

5.04

Score/$

Score: 75.6 · $15/1M

22
GPT-4o
OpenAI·Proprietary·128K

3.97

Score/$

Score: 39.7 · $10/1M

23
o3
OpenAI·Proprietary·200K

1.71

Score/$

Score: 68.3 · $40/1M

24
Claude Opus 4.6
Anthropic·Proprietary·1M

1.23

Score/$

Score: 92.4 · $75/1M

These rankings update weekly

Get notified when models move. One email a week with what changed and why.

Free. No spam. Unsubscribe anytime.

Key Takeaways

The best value model is Gemini 3.1 Flash-Lite by Google with a provisional Score/$ ratio of 98.75 (score: 39.5, output: $0.4/1M tokens).

The best open-weight model is DeepSeek Coder 2.0 at position #2.

24 models are included in this ranking.

Score in Context

What these scores mean

Value scores divide the weighted knowledge score by output token price (per 1M tokens). Higher means more capability per dollar. Models with no listed price are excluded.

Known limitations

Value rankings favor cheap models even if absolute performance is modest. A model scoring half as well at one-tenth the price wins on value — but may not meet your quality bar. Always check raw scores alongside value rankings.

Last updated: April 10, 2026

The AI models change fast. We track them for you.

For engineers, researchers, and the plain curious — a weekly brief on new models, ranking shifts, and pricing changes.

Free. No spam. Unsubscribe anytime.