Skip to content
← Models
Open weightsRenmin University of China/10 trustMetrics-checked

CodeS-15B

The SIGMOD 2024 open text-to-SQL model family: 1B to 15B, built specifically for schema-aware SQL generation.

Visit official page →

About

CodeS is a family of open text-to-SQL models from the RUCKBReasoning lab, published at SIGMOD 2024 in the paper 'Towards Building Open-source Language Models for Text-to-SQL'. It comes in 1B, 3B, 7B, and 15B sizes and is designed around the realities of real databases: schema linking, messy column names, and the kind of dirty data that breaks general-purpose chat models. Everything is Apache 2.0 licensed, with weights on Hugging Face under the seeklhy org.

The 15B model is the one to cite: the authors report 85.4% on Spider's dev set and 59.3% on the much harder BIRD benchmark, putting a small open model ahead of GPT-4 on Spider in their tests. Third-party evals have since confirmed it holds up well on standard benchmarks while collapsing, like everything else, on the newer enterprise-grade Spider 2.0 suite. For on-premise deployments that need SQL generation without sending schemas to a third party, it is one of the strongest fully open options.

Specs

Context window

Not published in surfaced materials

Pricing

Free / open weights (Apache 2.0)

Free quota

Free / open weights — self-host at no license cost

API access

Self-hosted; weights on Hugging Face (seeklhy/codes-15b); companion text2sql-demo repo

Weights

Open weights

Released

Not disclosed

Benchmarks

  • Spider dev: 85.4% (provider-reported; cited in the Spider 2.0 benchmark README)
  • BIRD: 59.3% (provider-reported)
  • Spider 2.0-lite: 0.7%, snow: 0.0% (third-party, Spider 2.0 repo — shows the enterprise gap)

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

text-to-sqlon-premise-analyticsdatabase-interfacesresearch

Trust score — 7/10

/10 trustMetrics-checked

Scored 7/10 (metrics tier, metrics-v1). Evidence: peer-reviewed SIGMOD 2024 paper (arXiv:2402.16347), Apache-2.0 licensed code and weights, numbers cited by independent benchmark repos. Strengths: academic rigor, permissive license, strong standard-benchmark results. Docked for: (1) headline Spider/BIRD numbers are author-run; (2) third-party Spider 2.0 results show the model fails on realistic enterprise workflows, which is the honest caveat to the strong scores.

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
7.0/10
Maintenance · recency, releases, issue load
6.0/10
Openness · license + docs
10.0/10
Manual review · sanity read + red-flag check
7.0/10
  • · SIGMOD 2024 peer-reviewed; numbers cited by independent benchmark repos
  • · Third-party Spider 2.0 results (0.7%/0.0%) expose the enterprise-reality gap

Sources