5 open source LLMs for coding and cost

This list ranks five open source LLMs by coding, reasoning, speed, context, and price.

Updated in April 2026, this ranking helps you pick from five open-weight models using live benchmark data, including a top Quality Index of 56.584.

Item	Quality Index	Best Price	Top Speed	Max Context
Qwen3.7 Max	56.584	$3.75/M	202 tok/s	991K
Kimi K2.6	53.905	$1.44/M	327 tok/s	262K
MiMo-V2.5-Pro	53.829	$1.20/M	88 tok/s	1M
DeepSeek V4 Pro	51.509	$0.54/M	159 tok/s	1M
MiniMax-M2.7	49.615	$0.52/M	446 tok/s	205K

1. [Qwen3.7 Max](https://whatllm.org/) for best overall quality

Qwen3.7 Max leads the open source ranking with a Quality Index of 56.584, which makes it the safest first pick if you want one model that can cover broad use cases without much guesswork. It also posts 202 tok/s and a 991K context window, so it is not just accurate, it is also practical for longer prompts.

• Best price: $3.75/M
• Top speed: 202 tok/s
• Max context: 991K
• Use it when you want the strongest all-around open-weight model in the list

Its main tradeoff is cost. If you are optimizing for budget or very high throughput, other models on this list give you more room to spend elsewhere. But for teams that care most about quality, Qwen3.7 Max is the benchmark to beat.

2. [Kimi K2.6](https://whatllm.org/) for speed and wide access

Kimi K2.6 sits near the top with a Quality Index of 53.905, but its bigger appeal is operational flexibility. It is the fastest model in this set at 327 tok/s and appears across 14 providers, which makes it easier to source and compare in real deployments.

• Best price: $1.44/M
• Top speed: 327 tok/s
• Max context: 262K
• Providers: 14

If you want an open-weight model that is easy to find and quick to respond, this is a strong choice. It is less attractive than Qwen3.7 Max for raw top-end quality, but it is better suited to apps where latency and provider choice matter more than absolute score.

3. [MiMo-V2.5-Pro](https://whatllm.org/) for long documents

MiMo-V2.5-Pro is the context monster in this ranking, with a 1M token window that can handle very large briefs, logs, or multi-file code sessions. Its Quality Index of 53.829 keeps it close to the leaders, so you do not have to give up much performance to get that extra room.

• Best price: $1.20/M
• Top speed: 88 tok/s
• Max context: 1M
• Good fit for document-heavy workflows

The tradeoff is speed. At 88 tok/s, it is not the quickest option here, so it is better for deep analysis than for chatty, high-volume traffic. Still, if your work lives in long prompts, this is one of the easiest models to justify.

4. [DeepSeek V4 Pro](https://whatllm.org/) for low cost at scale

DeepSeek V4 Pro gives you a strong balance of quality and price, with a 51.509 Quality Index and a very low listed price of $0.54/M. It also reaches 159 tok/s and supports a 1M context window, which makes it a flexible option for teams watching spend closely.

• Best price: $0.54/M
• Top speed: 159 tok/s
• Max context: 1M
• Good middle ground for production workloads

This is the model to watch if your priority is keeping API costs down without dropping into the lowest tier of performance. It is not the top scorer, but it is one of the most practical picks for sustained usage, especially when long context matters.

5. [MiniMax-M2.7](https://whatllm.org/) for the cheapest fast option

MiniMax-M2.7 is the budget speed pick, with the lowest listed price in this group at $0.52/M and the highest top speed at 446 tok/s. Its Quality Index of 49.615 is lower than the leaders, but it still stays competitive enough for many everyday tasks.

• Best price: $0.52/M
• Top speed: 446 tok/s
• Max context: 205K
• Providers: 6

Choose this when response time and cost matter more than squeezing out the last few benchmark points. It is especially appealing for high-volume chat, routing, or lightweight assistant use where fast turnaround is the real win.

How to decide

If you want the best overall open source model, start with Qwen3.7 Max. If you care most about speed and provider choice, Kimi K2.6 is the cleaner fit. For long documents, MiMo-V2.5-Pro is the easiest recommendation, while DeepSeek V4 Pro and MiniMax-M2.7 are the best value picks for lower spend.

For local or Ollama-style setups, hardware matters as much as model rank. In that case, use this list as a shortlist, then match the model to your VRAM, context needs, and tolerance for slower output.