Apply your own routing policies, reduce token costs automatically, and see every routing decision in real time with Merge Gateway.
What Devstral 2 costs to run
Test Devstral 2
with Gateway’s Simulator
See a prompt's output, token spend, latency, and more with Devstral 2.
Route requests to Devstral 2 in minutes
1$ pip install merge-gateway-sdk1from merge_gateway import MergeGateway
2
3client = MergeGateway(api_key="YOUR_API_KEY")
4
5response = client.responses.create(
6 model="openai/gpt-5.2",
7 input=[
8 {"type": "message", "role": "system", "content": "You are a helpful programming tutor. Explain the concepts clearly with practical examples."},
9 {"type": "message", "role": "user", "content": "Explain the concept of recursion in programming with a simple set of examples."},
10 ],
11)
12
13print(response.output[0].content[0].text)1response = client.responses.create(
2 model="anthropic/claude-sonnet-4-20250514",
3 input=[
4 {"type": "message", "role": "system", "content": "You are a helpful programming tutor. Explain the concepts clearly with practical examples."},
5 {"type": "message", "role": "user", "content": "Explain the concept of recursion in programming with a simple set of examples."},
6 ],
7)1from openai import OpenAI
2
3client = OpenAI(
4 api_key="YOUR_API_KEY",
5 base_url="https://api-gateway.merge.dev/v1/openai",
6)1response = client.chat.completions.create(
2 model="gpt-5.2",
3 messages=[
4 {"role": "system", "content": "You are a helpful programming tutor. Explain the concepts clearly with practical examples."},
5 {"role": "user", "content": "Explain the concept of recursion in programming with a simple set of examples."},
6 ],
7)
8
9print(response.choices[0].message.content)1npm install merge-gateway-ai-sdk-provider ai1import { createMergeGateway } from "merge-gateway-ai-sdk-provider";
2
3const gateway = createMergeGateway({
4 apiKey: "YOUR_API_KEY",
5});1import { generateText } from "ai";
2
3const { text } = await generateText({
4 model: gateway("openai/gpt-4o"),
5 prompt: "Explain the concept of recursion in programming with a simple set of examples.",
6});
7
8console.log(text);1import { createOpenAI } from "@ai-sdk/openai";
2
3const gateway = createOpenAI({
4 apiKey: "YOUR_API_KEY",
5 baseURL: "https://api-gateway.merge.dev/v1/ai-sdk",
6});
7
8// All generateText/streamText calls work unchanged1from anthropic import Anthropic
2
3client = Anthropic(
4 api_key="YOUR_API_KEY",
5 base_url="https://api-gateway.merge.dev/v1/anthropic",
6)
7
8message = client.messages.create(
9 model="claude-sonnet-4-20250514",
10 max_tokens=1024,
11 messages=[
12 {"role": "user", "content": "Explain the concept of recursion in programming with a simple set of examples."},
13 ],
14)
15
16print(message.content[0].text)Explore other models available in Merge Gateway
Devstral 2 FAQ
Heading
What other models does Mistral AI offer?
Mistral AI ships general-purpose models, coding-specialized models, and small open-weight models. Here are some other models Mistral AI supports:
- Mistral Medium 3.1: the mid-tier general-purpose model, balancing quality and cost for everyday text and multimodal tasks
- Mistral Large: the high-end general-purpose model for demanding reasoning and generation
- Mistral Small: a smaller, cheaper general model for high-volume tasks
How does Devstral 2 differ from Mistral AI's other models?
Devstral 2 is Mistral's coding-specialized, agentic model line, tuned for real software-engineering tasks rather than general chat.
- Coding focus: built for agentic coding and repository-scale work, reaching 72.2% on SWE-bench Verified, unlike the general-purpose Mistral models
- Context window: 256K tokens, suited to large codebases and long agent runs
- Sizes and licensing: ships as Devstral 2 (123B, modified MIT) and a smaller Devstral Small 2 (24B, Apache 2.0) light enough to run on a single high-end GPU or 32GB Mac
- Pricing: around $0.40 per 1M input and $2.00 per 1M output for Devstral 2, with the Small variant far cheaper
- Tooling: designed to pair with agentic coding harnesses, including Mistral's own Vibe CLI
Reach for Devstral 2 when your workload is code generation and agentic software tasks rather than general reasoning or chat.
What models should I consider using alongside Devstral 2?
No single model is optimal for every task. Here are models worth pairing with Devstral 2 depending on what your product needs:
- Claude Opus 4.8 for the hardest coding or architecture problems where you want frontier quality
- Gemini 3.1 Pro for long-context or multimodal steps a coding task hands off, like reading design docs
- Mistral Medium 3.1 for the general-purpose chat and reasoning steps around the coding work
- GPT-5.5 as a second high-quality general model for cross-provider comparison
What are the challenges of using Devstral 2 in my product?
Like any production LLM, Devstral 2 comes with tradeoffs worth planning for:
- Coding specialization: it's tuned for software tasks, so general chat, writing, or broad reasoning are better served by a general-purpose model
- Provider dependency: relying only on Mistral is fragile during outages or model retirements
- Self-host operations: running the open-weight version yourself means owning deployment, scaling, and upkeep
- Cost at scale: agentic coding loops make many calls, so budgeting and step limits matter even at Devstral 2's low rates
- Evaluate on your codebase: coding quality varies by language and repo structure, so test against your real code before committing
Why should I use Merge Gateway to route LLM requests with Devstral 2 and every other model?
Using Devstral 2 through Merge Gateway gives you access to the model itself and the infrastructure layer around it:
- One API, every provider: Reach Devstral 2 and every other major LLM through a single endpoint and API key, swapping the model string to change providers without touching application code
- Intelligent routing and automatic failover: Merge routes around Mistral outages automatically, and a Devstral-to-frontier quality ladder is easy to express as a policy that can cut spend 40 to 60%
- Cost governance: Set hard or soft project budgets so Devstral 2 spend stays in plan, with every request attributed to a model, project, and tag in one billing dashboard
- Build Your Own Router: Define what "best" means with curated benchmarks or your own coding evals, and the router picks per request and explains why
- Security and compliance controls: Apply DLP rules and prompt injection protection before requests reach Mistral, and enforce per-project model and region policies outside your application
How can I start routing requests to Devstral 2 via Merge Gateway?
Getting Devstral 2 running through Merge Gateway takes a few minutes:
1. Create an account and get your API key from the dashboard.
2. Install the Merge Gateway SDK: run pip install merge-gateway-sdk (Python) or npm install merge-gateway-sdk (Node). Alternatively, if you're already using the OpenAI SDK, set base_url = "https://api-gateway.merge.dev/v1/openai" and your existing code works as-is.
3. Make your first request using the provider/model format. For Devstral 2, the model string is mistral/devstral-2 (confirm the exact slug in the dashboard). Swap the model string to route to any other provider without changing anything else.
4. Configure a routing policy in the dashboard to set failover behavior, cost limits, and optimization strategy. Your first policy can be as simple as naming Devstral 2 as primary with one fallback.
Full setup instructions and SDK references are in the Merge Gateway docs.
Try Devstral 2 through Merge Gateway
Route, observe, and control AI requests across providers from one API.
