Llama 3.1 8B Instruct, Mistral 7B Instruct v0.3, and Qwen 3 8B Instruct are the canonical sub-10B options that teams benchmark before deciding whether to step up to the 14B or 32B tier. All three are instruction-tuned, broadly hosted, and priced at the cost floor for practical general-use inference. Mistral 7B Instruct v0.3, released May 2024, is the oldest of the three and the baseline most teams use as a price anchor. It added native function-calling and an extended vocabulary over earlier Mistral 7B releases. Pricing typically sits below $0.10 per million tokens across providers. The 32K context window covers most summarization and classification workloads. Apache 2.0 license makes it frictionless for fine-tuning and redistribution. Meta's Llama 3.1 8B Instruct, released July 2024 under the Llama 3 community license, is the sub-10B model with the widest provider coverage. The 131K context window is its clearest structural advantage over Mistral 7B — four times the context headroom for the same inference cost tier. On general instruction-following and coding benchmarks it sits ahead of Mistral 7B v0.3. New deployments building on the Llama ecosystem have a clear reason to use this unless Apache 2.0 is a hard legal requirement. Qwen 3 8B Instruct from Alibaba brings the strongest multilingual coverage of the three, with measurably better CJK and Arabic handling plus a 131K context window that matches Llama 3.1 8B. On reasoning benchmarks, Qwen 3 8B competes closely with Llama 3.1 8B for the top position in this size class. The Qwen license permits commercial use with attribution. Pick Mistral 7B v0.3 when Apache 2.0 is required and context needs are modest. Pick Llama 3.1 8B for maximum provider optionality and long-context use cases. Pick Qwen 3 8B when multilingual breadth or CJK performance is the deciding factor.