Amazon is cranking up the pace of its AI rollout, adding 18 new open-weight models to Amazon Bedrock in one of its biggest expansions yet.
The update drops new multimodal power into the platform and brings Mistral’s newest models on board first.
In a post on the AWS News Blog, lead blogger Channy Yun said the expansion pushes Bedrock toward a nearly 100-model lineup, with fresh entries from Mistral, Google, Nvidia, OpenAI, MiniMax, Moonshot, and Qwen.
AWS cast the refresh as a broad upgrade aimed at giving customers more choice without touching their existing infrastructure.
A broad mix of AI providers joins the latest expansion
The new release adds a varied set of models from leading AI developers. Among the standout additions:
- Google Gemma 3: Lightweight multimodal models for local text-and-image work on laptops and workstations.
- Nvidia Nemotron Nano 2: Efficiency-focused models built for reasoning, coding, and video understanding.
- OpenAI gpt-oss-safeguard: New safety classifiers for policy enforcement and high-volume content moderation.
- MiniMax M2: Coding-automation models capable of multi-file edits and long tool-calling chains.
- Moonshot Kimi K2 Thinking: A deep reasoning model for research-heavy, multistep workflows.
- Qwen3-Next and Qwen3-VL: Long-context and vision models supporting document extraction, code generation, and video analysis.
The mix spans language, vision, audio, and safety workloads, collectively expanding Bedrock’s options as it moves toward a triple-digit model catalog.
Mistral models take the spotlight
Mistral’s newest lineup is the centerpiece of the update.
Mistral Large 3 steps in as a long-context, multimodal workhorse built for dense enterprise tasks, from document-heavy workflows and multilingual analysis to advanced coding and tool-using agents. It’s tuned for reliability on lengthy prompts and for reasoning that spans both text and vision.
Alongside it, AWS is rolling out the full Ministral 3 range — 3B, 8B, and 14B — all designed to run efficiently on a single GPU. The smallest model targets lightweight vision and language tasks on edge devices, while the 8B variant balances footprint and performance for chat interfaces and embedded systems. The 14B version caps the set with state-of-the-art text and vision capabilities for private, on-premises deployments where hardware headroom is tight.
AWS says Bedrock is the first platform to host these models, giving customers early access to Mistral’s latest long-context and edge-optimized releases.
Stronger audio, vision, and reasoning features
The rest of the lineup brings in capabilities that don’t often land in a single release.
New audio models handle speech with more precision, from rapid transcription to multilingual voice commands that work without a constant cloud connection.
Vision-focused systems enhance how software reads documents, turns screenshots into working code, and interprets video sequences with far more context.
Others lean into problem-solving, such as long-form planning, multi-step tool use, and retrieval workflows that can keep track of sprawling inputs. The result is a set of models that can slot into everything from on-device processing to moderation queues, automation tools, and industry workflows that depend on richer multimodal understanding.
A clearer ramp for testing and deploying the latest models
AWS is rolling out the support stack with the AI models, making it easier for teams to try them out and move them into production. The unified API remains the entry point, so swapping or comparing models doesn’t require reworking an application.
Developers can test everything in the Bedrock console’s playground or wire the models directly into their systems using AWS SDKs. For teams building agents, frameworks like Bedrock AgentCore and Strands Agents are already set up to work with the new releases.
AWS is expanding the evaluation and safety toolkit as well. Guardrails can be applied to any incoming model, and the built-in evaluation suite helps teams benchmark options before choosing one for a workload.
With the update live, AWS now enters its next phase: scaling access as customers start testing the new models.
Lyft’s latest collaboration with Anthropic and AWS offers a glimpse of agentic AI stepping into large-scale customer support.