whatmodel.sh / Solar Open 2 (250B-A15B)

Solar Open 2 (250B-A15B)

+ contribute

Solar Open 2 (250B-A15B) is an open-weights language model from Upstage, released July 2026. Its published size is 250.29B parameters (15B active per token) and a 1M-token context window. Licence: upstage-solar-license.

Catalogue identity notes

Upstage's own open-weights model, repo upstage/Solar-Open2-250B @9190fbe (created 2026-07-22, commit 'Release Solar Open2 250B'; licence tag other / upstage-solar-license; safetensors metadata 250,287,810,304 parameters). The technical report's column 'Solar Open 2 / 250B-A15B' is this checkpoint (the card links the same report and prints the same Table 3). No first-party API page verified; local vLLM only.

Model information sources

Release date ↗ · Weights licence ↗

Model card (upstage/Solar-Open2-250B @9190fbe): 'Total Parameters | 250B (250,287,794,944)', 'Active Parameters | 15B (per token)'; report Table 1 '250B' / '15B'.

71.6±1.3#70 of 719 LLMsfrom 13 benchmark results across 1 source

$ how it compares

The LLMs ranked either side of Solar Open 2 (250B-A15B), out of 828 with a score, then more from Upstage.

#	model	score	where	RAM	released
73	DeepSeek V4 Flash (Preview, Apr 2026) DeepSeek	72±1	local	~103GB	2026-04
74	Grok 4.20 (0309 reasoning) SpaceXAI (xAI)	72±1	api $1.56/M		2026-04
75	Solar Open 2 (250B-A15B) Upstage	72±1	local	~527GB	2026-07
76	North Small Translate 1.0 Cohere	71±6	api price unknown		2026-08
77	GPT-5.4 mini OpenAI	71±1	api $1.69/M		2026-03
	more from Upstage
170	Solar Pro 4 Upstage	65±2	api $0.53/M		2026-08
240	Solar Open 100B (102B-A12B) Upstage	60±1	local	~41GB	2025-12
662	SOLAR 10.7B Instruct v1.0 Upstage	37±1	local	~8.3GB	2023-12

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #63 of 300 · 30% of its score · 5 tests

Beats Gemini 3.6 Flash, GPT-5 Pro, LongCat 2.0 (1.6T-A48B) here, though they rank higher overall. Trails MiniMax-M3, GLM-5.1, Mistral Large 4 (public preview) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
APEX-Agents (pass@1) lab	16.6% ✎	▄▄····▄···█▄ beats 5/5	very hard · avg 1.8%	▇▇▇▇ 10%	typical
Terminal-Bench Hard lab	28.3% ✎	▄········██▄ beats 2/5	very hard · avg 4.5%	▇▇▇▇ 9.1%	sharp · at its level
MCP-Atlas (public set) lab	58.2% ✎	▄▄▄·······█▄ beats 3/5	hard · avg 19%	▇▇ 5.5%	at its level
SWE-bench Verified lab	70.4% ✎	▄··········█ beats 3/5	hard · avg 16%	▇ 3.5%	sharp · at its level · noisy
τ³-bench Banking lab	19.6% ✎	█·▄▄······▄▄ beats 4/5	very hard · avg 5.1%	▇ 2.2%	blunt

coding▁▁▁▃▆██▇▆▄▃▁ #62 of 458 * · 8.1% of its score · 2 tests

Beats DeepSeek V4 Flash Vision (Exp), Gemini 3.1 Pro (Preview), GPT-5.2 here, though they rank higher overall. Trails MiMo-V2.5-Pro (1T-A42B), MiniMax-M3, Ornith 1.5 35B-A3B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LiveCodeBench v6 (window unspecified) lab	92.4% ✎	▄······▄▄▄·█ beats 5/5	medium · avg 50%	▇▇ 4.6%	typical
SWE-bench Verified lab	70.4% ✎	▄··········█ beats 3/5	hard · avg 16%	▇ 3.5%	sharp · at its level · noisy

math▁▂▃▅▆▅▄█▄▁▁▁ #59 of 398 * · 11% of its score · 2 tests

Beats Qwen3.6 Max Preview, Claude Sonnet 4.6, Claude Opus 4.5 here, though they rank higher overall. Trails INTELLECT-3, GLM-4.7, A.X K2 (688B-A33B) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
AIME 2026 lab	95.7% ✎	▄▄··▄····█·▄ beats 3/5	medium · avg 64%	▇▇▇ 7.2%	sharp · maxed out
HMMT February 2026 lab	93.9% ✎	█▄▄········█ beats 4/5	hard · avg 31%	▇▇ 4.0%	sharp · noisy

other tests5 not in a task · 53% of its score

test	result	vs the others in its run	difficulty	counts	why
GPQA Diamond lab	86.3% ✎	█···██·█··██ beats 4/5	hard · avg 42%	▇▇▇▇▇▇▇▇ 21%	sharp
Humanity's Last Exam lab	28.8% ✎	█▄······▄·▄▄ beats 4/5	very hard · avg 4.7%	▇▇▇▇▇▇ 14%	at its level
MMLU-Pro lab	86.2% ✎	▄·▄▄····▄··█ beats 5/5	medium · avg 55%	▇▇▇ 7.4%	typical
IFBench (aggregation unspecified) lab	80.0% ✎	▄···▄▄··▄··█ beats 4/5	medium · avg 43%	▇▇ 5.7%	typical
MultiChallenge lab	61.0% ✎	█··▄·▄·····█ beats 4/5	hard · avg 25%	▇▇ 5.3%	at its level

$ where to run it

Other ways to run it

Run it as published · needs ~500.58 GB, NVIDIA only · from huggingface.co/upstage (official)
vllm serve upstage/Solar-Open2-250B --served-model-name solar-open2-250b --tensor-parallel-size 8 --enable-expert-parallel --moe-backend triton --default-chat-template-kwargs '{"think_render_option":"preserved"}' --reasoning-parser solar_open2 --tool-call-parser solar_open2 --enable-auto-tool-choice --logits-processors vllm.v1.sample.logits_processor.solar_open2:SolarOpen2TemplateLogitsProcessor

Official weights and documentation: huggingface.co/upstage ↗

$ how the score was worked out · Solar Open 2 Technical Report… 99%

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

source	results	share of the score	licence
Solar Open 2 Technical Report (Upstage, arXiv:2607.20062v2) · lab-reported	13	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 99%	CC-BY-4.0
prior · pull toward the middle		▇ 1%
weighted average		= 71.6

$ improve this page

Update model information Add a place to run it Add a test result Propose a benchmark Correct a benchmark score Edit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated 2026-10-11 02:20 UTC · sources and licences · json