Models

A smaller, cheaper member of the GPT-5.1 family that keeps much of the reasoning quality at a fraction of the price. Configurable reasoning effort and strong instruction-following make it a popular middle ground for production assistants that need more than a budget open model but can't justify flagship pricing. First-party hosted.

Mistral's dedicated code model. Fill-in-the-middle completion, low-latency IDE autocomplete, dozens of languages. Cheap and fast for quick code tasks. Not a general chat model; use it when the workload is specifically coding.

MiniMax's efficient long-context model with a launch-priced cheap input tier and a million-token context window. Strong at long-document reasoning and retrieval, increasingly used for agents that need to hold a lot of state. Hosted with growing provider coverage.

Alibaba's open coding model, tuned for repository-scale code generation and agentic tool use, with a very long context window. Among the strongest open options for coding at a low price, and widely hosted. Overkill for plain text classification, but excellent when the task is actually writing or editing code.

Google's workhorse hosted model: fast, multimodal and inexpensive, with a very large context window. A step up from Flash-Lite in reasoning and coding while staying cheap enough for high-volume use. First-party served, and a common default for production chat, summarization and light agentic work.

DeepSeek's full V4 flagship, a large MoE with strong reasoning, coding and math, released under a permissive license. It competes with first-party frontier models on hard benchmarks at a fraction of the price, which has made it a favorite for cost-sensitive teams running complex work. Multiple providers host it, so LLMRouteHub can route…

Baidu's flagship multimodal model, strong on Chinese-English tasks, reasoning and tool use. Hosted with solid uptime and a large context window. A capable option for bilingual production workloads; priced in the mid-tier rather than for bulk jobs.

DeepSeek's open reasoning model, the successor to R1, with long chain-of-thought and strong math, coding and planning. Competitive with the best hosted reasoners at a fraction of the cost, though reasoning tokens add to output spend. A favorite for hard problems where you still care about price.

Z.ai's open-weight GLM-5, a large model tuned for agentic coding and tool use that has become popular as a low-cost backbone for autonomous coding agents. Strong real-world coding performance for the price, hosted by several providers under a permissive license. A good open alternative to the pricier hosted agent models.

Moonshot AI's open-weight Kimi K2.5, a large MoE tuned for agentic workflows, long-document reasoning and tool use. Competitive with much pricier hosted models on coding and multi-step tasks, and increasingly popular for autonomous agents. Available from several providers under a permissive license, with a big context window for…

Amazon's mid-tier hosted model, served on Bedrock. Solid general-purpose quality, multimodal input, dependable enterprise uptime, tight AWS integration. A reasonable managed choice if you're already on AWS. Priced above the open mid-tier.

Anthropic's small, fast hosted model. Excellent instruction-following and strong safety behavior make it a popular choice for customer-facing assistants and moderation. Snappy latency and dependable uptime, first-party served only. It's priced well above the open small models, so it tends to win on quality-per-call rather than raw cost…