Qwen2.5 14B Instruct vs Aya Expanse 32B vs Codestral

Codestral is the better all-round choice, scoring 48/100 against Qwen2.5 14B Instruct (42) and Aya Expanse 32B (33). It leads on price and context window. The score weighs price 50%, inputs & features 30%, context window 20%. None of the shared benchmarks cover every model here yet, so capability is left out of this verdict.

Codestral is cheaper at $0.30 input / $0.90 output per million tokens (official Mistral API price). Qwen2.5 14B Instruct costs $0.35 input / $1.40 output per million tokens (official Alibaba API price); Aya Expanse 32B costs $0.50 input / $1.50 output per million tokens (median across 1 API provider). At a typical mix of three input tokens to one output token, that is $0.45 per million tokens for Codestral versus $0.613 for Qwen2.5 14B Instruct (1.4× as much) and $0.75 for Aya Expanse 32B (1.7× as much).

There is no independent benchmark that covers all three models yet. Qwen2.5 14B Instruct has not been scored yet, Aya Expanse 32B has not been scored yet and Codestral has not been scored yet.

There are no published SWE-bench Verified results for Qwen2.5 14B Instruct, Aya Expanse 32B and Codestral yet, so there is no like-for-like coding score. Test both on a sample of your own tasks. Note that Aya Expanse 32B does not support tool calling, which most coding agents need.

Codestral has the largest context window at 256,000 tokens, against 131,072 for Qwen2.5 14B Instruct and 128,000 for Aya Expanse 32B. Maximum output per response: Qwen2.5 14B Instruct up to 8,192, Aya Expanse 32B up to 4,000, Codestral up to 4,096 tokens.

Qwen2.5 14B Instruct accepts text; Aya Expanse 32B accepts text; Codestral accepts text. They handle the same number of input types.

Yes, all three publish their weights (CC-BY-NC-4.0), so you can self-host them.

Aya Expanse 32B is the newest, released Oct 24, 2024. Qwen2.5 14B Instruct came out Sep 2024; Codestral came out May 29, 2024. Knowledge cutoff: Qwen2.5 14B Instruct Apr 2024, Codestral Oct 2024.

Each model gets a 0–100 score on capability (50%, independent benchmark results); price (25%, blended price per million tokens (3 input : 1 output), log scale); inputs & features (15%, image, PDF, audio and video input, tool calling, structured output and reasoning); context window (10%, maximum tokens per request, log scale). Dimensions missing for any model are dropped and the remaining weights rescaled, so every model is judged on the same evidence. Specs and prices come from public model listings and the labs’ own API pages; capability scores come from independent benchmark runs. Data updated Oct 4, 2026.