Solar Pro 2
Upstage's Korean-centric frontier LLM: Korea's first to outrank GPT-4.1 on a global frontier index.
About
Solar Pro 2 is the July 21, 2025 flagship of Korean startup Upstage, and the model that briefly made Korea a frontier-model country: it scored 58 on the Artificial Analysis Intelligence Index, five points above GPT-4.1 and ahead of DeepSeek V3 and Kimi K2 — the first Korean model to crack the global frontier top tier, per KoreaTechDesk's reporting of the AA data. It is a Korean-centric enterprise LLM aimed at document-heavy, high-stakes work in finance, healthcare and legal, continuing Upstage's single-GPU-efficiency lineage.
Solar Pro 2 itself is a proprietary API model served through the Upstage console — Upstage's openness play went instead to Solar Open 2 (July 2026), a 250B-A15B hybrid-attention MoE with a 1M-token context released as open weights under the Apache-based Upstage Solar License. For teams that need Korean-language frontier performance with a vendor API rather than self-hosting, Solar Pro 2 remains Upstage's commercial flagship.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- Artificial Analysis Intelligence Index: 58 (third-party, Jul 2025 — above GPT-4.1 at 53, DeepSeek V3 at 53, and Kimi K2 at 57.6, per KoreaTechDesk citing AA)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Upstage release Jul 21, 2025; third-party Artificial Analysis Intelligence Index score of 58 above GPT-4.1 (reported via KoreaTechDesk citing AA); vendor actively maintained the line through Solar Pro 3/4 and open-weighted Solar Open 2 (Jul 2026). Closed weights, API-only. Release facts corroborated across KoreaTechDesk, Upstage's blog, and independent licensing reviews. Strengths: credible third-party frontier ranking, Korean-language specialization, active vendor. Docked for: (1) parameter count and context window not published by Upstage — core specs unverifiable; (2) closed weights with all capability claims resting on a single third-party index snapshot.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 6.0/10
- Maintenance · recency, releases, issue load
- 8.0/10
- Openness · license + docs
- 3.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Upstage release 2025-07-21; AA Intelligence Index 58 (third-party) above GPT-4.1 (53); vendor active through Solar Pro 3/4 + open Solar Open 2
- · Docked: Parameter count and context window not published — core specs unverifiable
- · Docked: Closed weights, API-only; claims rest on a single third-party index snapshot