Upstage Solar is a three-model lineup from Upstage, a Korean lab whose Solar API prices input and output identically. Driving Atlas with Solar means solar-pro3 at $0.25/$0.25 per Mtok across 131,072 tokens, solar-pro2 at the same $0.25/$0.25 across 65,536 tokens, and solar-mini at $0.15/$0.15 across 32,768 tokens. Symmetric pricing is rare and it changes the math: a verbose agent turn costs the same per token as the context you fed it. The lineup has no frontier tier, and output caps at 8,192 tokens on both Solar Pro models.
What models are in the Upstage Solar lineup?
The Upstage Solar lineup is three models. Solar Pro 3 holds 131,072 tokens at $0.25/$0.25 per Mtok. Solar Pro 2 holds 65,536 tokens at the same $0.25/$0.25. solar-mini holds 32,768 tokens at $0.15/$0.15 per Mtok.
Upstage keeps its catalog deliberately small, and all three Solar models share the same symmetric price shape. Solar Pro 3 and Solar Pro 2 cost an identical $0.25 per Mtok in both directions, so the only thing separating them is the window: 131,072 tokens versus 65,536. solar-mini sits below both at $0.15/$0.15 per Mtok with a 32,768 token window. Upstage is a lab that actually tunes for Korean and for document extraction, which is a narrower specialization than the general labs pursue. Atlas lets you switch the active model and provider on the fly with favorites and recents, so carrying all three Solar entries and moving between them is a normal way to work rather than a reconfiguration.
Which Upstage Solar model belongs in each Atlas slot?
In Atlas, pin `"model": "upstage/solar-pro3"` and `"small_model": "upstage/solar-mini"` in atlas.json. Solar Pro 3 at $0.25/$0.25 per Mtok drives the work, and solar-mini at $0.15/$0.15 per Mtok absorbs the overhead calls.
Atlas runs two model slots, and the Upstage Solar lineup fills both from one provider. Solar Pro 3 takes the main slot with its 131,072 token window. solar-mini takes the cheap slot, and at $0.15/$0.15 per Mtok with a 32,768 token window it is a cheap, dedicated `small_model` from a lab that actually tunes for Korean and for document extraction. Keeping both slots on Upstage means one API key, one endpoint, and one billing relationship. Atlas is a terminal-native TUI rendered with SolidJS through the OpenTUI renderer, and Atlas ships a TUI theme system with a charcoal-and-blue default theme and many presets, so the interface stays the same whichever Solar model is active.
Should you move from Solar Pro 2 to Solar Pro 3?
Yes. Solar Pro 3 doubles the context of Solar Pro 2, 131,072 versus 65,536 tokens, while keeping the exact same price of $0.25/$0.25 per Mtok. There is no cost argument for staying on Solar Pro 2 once Solar Pro 3 is available on the same Upstage API key.
The Solar Pro 2 to Solar Pro 3 upgrade is unusually clean, because Upstage did not reprice the larger window. Both models bill at $0.25 per Mtok in and $0.25 per Mtok out, so doubling from 65,536 to 131,072 tokens of context costs nothing extra per token. In Atlas the migration is a single string in atlas.json. Run `atlas models upstage` to confirm the three Solar models resolved, then change the pinned entry. Atlas can build its code index with local Ollama embeddings, keeping code off third-party servers, so the amount of your codebase that reaches Upstage at all is a separate decision from which Solar model you pin.
What are the output limits on Upstage Solar?
Max output is 8,192 tokens on both Solar Pro models and 4,096 on solar-mini, so large single-shot file rewrites will not fit. That ceiling is the defining constraint on the Upstage Solar lineup, and it holds regardless of the 131,072 token context on Solar Pro 3.
A large context window and a small output ceiling is a specific shape: Upstage Solar reads widely and writes narrowly. Solar Pro 3 can take in 131,072 tokens and hand back at most 8,192. solar-mini takes in 32,768 and hands back at most 4,096. The practical guidance is to keep changes scoped and lean on Atlas's unified-diff review rather than whole-file rewrites. Atlas connects to Model Context Protocol servers and exposes their tools to the agent, and Atlas is extensible through plugins that contribute tools and hook into agent lifecycle events, so work that would otherwise be emitted as a giant blob of generated text can often be done through a tool call instead.
When is Upstage Solar the wrong choice for a coding agent?
Upstage Solar is the wrong choice when you need a frontier tier. Upstage ships three models total. There is no frontier tier here, and Solar has not been benchmarked head to head with Claude or GPT-5 on large-repo agentic coding, so a hard refactor is not where Solar has earned trust.
Two facts bound the Upstage Solar lineup honestly. The catalog is three models deep, so there is nothing above Solar Pro 3 to escalate to inside Upstage. And Solar carries no published head to head against Claude or GPT-5 on large-repo agentic coding, which means adopting it is a decision you make on your own evaluation rather than on someone else's leaderboard. Atlas softens that by making the switch cheap. Atlas lets you switch the active model and provider on the fly with favorites and recents, so you can run Solar Pro 3 at $0.25/$0.25 per Mtok for cost-sensitive work and escalate to another provider for the tasks Solar has not proven itself on.
Setup
- 01Export UPSTAGE_API_KEY. Atlas resolves https://api.upstage.ai/v1/solar via @ai-sdk/openai-compatible.
- 02Run `atlas models upstage` to confirm the three Solar models resolved.
- 03Pin `"model": "upstage/solar-pro3"` and `"small_model": "upstage/solar-mini"` in atlas.json.
- 04Because output is capped at 8,192 tokens, keep changes scoped and lean on Atlas's unified-diff review rather than whole-file rewrites.
- 05Add the Solar entries to favorites so you can move between solar-pro3 and solar-mini without editing atlas.json again.
Frequently asked questions
- what models does upstage solar offer
- Upstage Solar is three models: solar-pro3 with a 131,072 token window, solar-pro2 with 65,536 tokens, and solar-mini with 32,768 tokens. Solar Pro 3 and Solar Pro 2 both cost $0.25/$0.25 per Mtok; solar-mini is $0.15/$0.15.
- how much does upstage solar cost per million tokens
- solar-pro3 is $0.25/$0.25 per Mtok, solar-pro2 is $0.25/$0.25, and solar-mini is $0.15/$0.15 per Mtok. Upstage's Solar API prices input and output identically, which is unusual across the registry.
- solar pro 2 vs solar pro 3 which should I use
- Solar Pro 3. It doubles the context of Solar Pro 2, 131,072 versus 65,536 tokens, while keeping the exact same $0.25/$0.25 per Mtok price, so there is no cost reason to stay on Solar Pro 2.
- how do I connect atlas to the upstage api
- Export UPSTAGE_API_KEY. Atlas resolves https://api.upstage.ai/v1/solar via @ai-sdk/openai-compatible. Run `atlas models upstage` to confirm the three Solar models resolved, then pin `"model": "upstage/solar-pro3"` in atlas.json.
- what is the max output of upstage solar models
- Max output is 8,192 tokens on both Solar Pro models and 4,096 on solar-mini. Large single-shot file rewrites will not fit, so scope changes small and review the diff instead.
- does upstage have a frontier model
- No. Upstage ships three Solar models total. There is no frontier tier, and Solar has not been benchmarked head to head with Claude or GPT-5 on large-repo agentic coding.
- cheapest upstage model for the atlas small_model slot
- solar-mini at $0.15/$0.15 per Mtok with a 32,768 token window. Set `"small_model": "upstage/solar-mini"` in atlas.json to keep both Atlas slots on one Upstage API key.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Atlas for Node.js in 2026
Node.js developers in 2026 can adopt Atlas, the terminal-native AI coding agent, for secure, efficient development. Leverage local embeddings, AST indexing, and robust safety features.
Atlas for Nim: A Terminal-Native AI Coding Agent for Nimble Packages and Macros in 2026
Atlas is a terminal-native AI coding agent for Nim in 2026. It reads .nimble requires and asterisk-exported symbols, adds std/unittest suites, runs nimble test, formats with nph.
Atlas for Haskell in 2026
Atlas is a terminal-native AI coding agent for Haskell in 2026. Run it in a project with a .cabal file or a stack.yaml, let cabal build drive it, and review the diff.
Atlas for Gin in 2026
Atlas is a terminal-native AI coding agent for Gin in 2026. It reads router groups and binding tags, then runs go test ./... -race behind a permission prompt.
Atlas vs OpenAI Codex CLI: Terminal AI Coding Agents in 2026
Comparing Atlas and OpenAI Codex CLI for terminal AI coding in 2026. Atlas offers a TUI, permission-gated tools, and BYO models, while OpenAI Codex CLI provides sandboxed execution and OpenAI model integration.
Atlas for Fiber in 2026
Atlas is a terminal-native AI coding agent for Fiber in 2026. It knows fasthttp reuses buffers, tests handlers with app.Test(), and diffs every edit first.
Atlas vs Cosine: Terminal AI Coding Agents in 2026
Comparing Atlas and Cosine for terminal AI coding in 2026. Atlas offers a free core and local privacy, while Cosine provides proprietary models and a cloud surface.
Atlas vs Trae: AI Coding Agents in 2026
Atlas and Trae offer distinct AI coding agent experiences in 2026. Atlas provides terminal-native control and local data privacy, while Trae offers a full IDE with SOLO Builder and cloud tasks.