Images as input
Choose Grok 4.20 Non-Reasoning
Grok 4.20 Non-Reasoning accepts images as input; Devstral 2 does not.
Devstral 2 vs Grok 4.20 Non-Reasoning: price, context and capabilities
Choose Grok 4.20 Non-Reasoning for images as input and the latest release; on the other points the two are level or the data is missing.
In short
- Devstral 2 has been retired; Grok 4.20 Non-Reasoning is the one you can still call.
- A price comparison is not possible: one of them has no comparable published price.
- Grok 4.20 Non-Reasoning is the newer model (March 2026, against December 2025).
- Only Grok 4.20 Non-Reasoning offers vision, tool calling, structured output, prompt caching, batch api.
Which should you choose?
Each answer below comes straight from the published prices, specifications and independent benchmark results.
Latest release
Choose Grok 4.20 Non-Reasoning
Grok 4.20 Non-Reasoning is the newer release (March 2026, against December 2025 for Devstral 2).
Specifications side by side
| Devstral 2 | Grok 4.20 Non-Reasoning | |
|---|---|---|
| Provider | Mistral AI | xAI |
| Category | Language models | Language models |
| Status | Retired | Generally available |
| Headline price | No data | $1.56 / 1M tokens |
| Input | No data | $1.25 / 1M tokens |
| Output | No data | $2.50 / 1M tokens |
| Cached input | No data | $0.20 / 1M tokens |
| Context window | No data | 1M |
| Max output | No data | No data |
| Released | December 2025 | March 2026 |
| Knowledge cutoff | No data | No data |
| Licence | Proprietary | Proprietary |
| Capabilities | No data | Vision, Tool calling, Structured output, Prompt caching, Batch API |
| API model id | devstral-2512 | grok-4.20-0309-non-reasoning |
Benchmarks side by side
Leader in bold.
| Benchmark | Devstral 2 | Grok 4.20 Non-Reasoning |
|---|---|---|
| Terminal-Bench Hard | No data | 16.7% |
| GPQA Diamond | No data | 77.6% |
| Humanity's Last Exam | No data | 27.9% |
| IFBench (instruction following) | No data | 49.3% |
| τ²-Bench (agentic tool use) | No data | 59.9% |
| MMMU-Pro (visual reasoning) | No data | 64.9% |
| Long-context reasoning (AA-LCR) | No data | 23.3% |
Independent evaluations by Artificial Analysis, run the same way for every model, at the reasoning setting shown. Percentages unless marked as an index.
Frequently asked questions
Should I choose Devstral 2 or Grok 4.20 Non-Reasoning?
Choose Grok 4.20 Non-Reasoning for images as input and the latest release; on the other points the two are level or the data is missing.
Which is cheaper, Devstral 2 or Grok 4.20 Non-Reasoning?
A price comparison is not possible: one of them has no comparable published price.
What is the difference between Devstral 2 and Grok 4.20 Non-Reasoning?
Devstral 2 is made by Mistral AI and Grok 4.20 Non-Reasoning by xAI. Devstral 2 has been retired; Grok 4.20 Non-Reasoning is the one you can still call. A price comparison is not possible: one of them has no comparable published price. Grok 4.20 Non-Reasoning is the newer model (March 2026, against December 2025). Only Grok 4.20 Non-Reasoning offers vision, tool calling, structured output, prompt caching, batch api.
Which is newer, Devstral 2 or Grok 4.20 Non-Reasoning?
Grok 4.20 Non-Reasoning was released on March 2026 and Devstral 2 on December 2025.