SpringPrompt predicted fit · public beta
Highest-throughput AI models for Business Writing
Predicted ability to produce clear, coherent and audience-appropriate business prose such as briefs, updates, proposals and memos.
gpt-oss-120b has the highest source-listed output throughput among business writing models with an active operational reference.
This is a task-specific prediction from several external benchmark signals. It is not a task-success percentage, guarantee, strict league table, or fresh direct model run. Close scores should be read as the same performance band.
Highest output throughput (beta)
Active reference configurations in the same evidence-qualified task cohort are ranked by source-listed output tokens per second. Deprecated references remain visible as legacy data without a throughput position. This is generation throughput after output begins—not end-to-end response speed—and it does not change predicted fit.
Updated 2026-07-16
| Highest output throughput | Model family | Predicted fit | Blended price† | Output throughput† | Evidence coverage | Independent evidence |
|---|---|---|---|---|---|---|
| #1fit #30 | gpt-oss-120b openai Operational reference: gpt-oss-120b (high) ↗Active configuration | 45.5relative task estimate | $0.20per 1M blended | 255.3 tok/sexcludes first-token wait | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| #2fit #31 | gpt-oss-20b openai Operational reference: gpt-oss-20b (high) ↗Active configuration | 43.0relative task estimate | $0.07per 1M blended | 205.4 tok/sexcludes first-token wait | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| #3fit #15 | Grok 4.5 xai Operational reference: Grok 4.5 (high) ↗Active configuration | 57.0relative task estimate | $1.35per 1M blended | 116.3 tok/sexcludes first-token wait | 55% | 3 signal groups · 3 lineages EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| #4fit #18 | Gemini 3.1 Pro google Operational reference: Gemini 3.1 Pro Preview ↗Active configuration | 56.0relative task estimate | $1.74per 1M blended | 115.9 tok/sexcludes first-token wait | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| #5fit #16 | DeepSeek-V4-Flash deepseek Operational reference: DeepSeek V4 Flash (max) ↗Active configuration | 56.9relative task estimate | $0.06per 1M blended | 108.6 tok/sexcludes first-token wait | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| #6fit #9 | Claude Sonnet 5 anthropic Operational reference: Claude Sonnet 5 (max) ↗Active configuration | 59.2relative task estimate | $1.54per 1M blended | 70.9 tok/sexcludes first-token wait | 55% | 3 signal groups · 3 lineages EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| #7fit #2 | GPT 5.5 openai Operational reference: GPT-5.5 (xhigh) ↗Active configuration | 61.8relative task estimate | $4.35per 1M blended | 66.3 tok/sexcludes first-token wait | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| #8fit #3 | Claude Fable 5 anthropic Operational reference: Claude Fable 5 (with fallback) ↗Active configuration | 61.6relative task estimate | $7.70per 1M blended | 65.0 tok/sexcludes first-token wait | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| #9fit #12 | Deepseek V4 Pro deepseek Operational reference: DeepSeek V4 Pro (max) ↗Active configuration | 57.8relative task estimate | $0.18per 1M blended | 62.3 tok/sexcludes first-token wait | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| #10fit #21 | Qwen3.5-397B-A17B alibaba Operational reference: Qwen3.5 397B A17B ↗Active configuration | 53.1relative task estimate | $0.90per 1M blended | 59.6 tok/sexcludes first-token wait | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| #11fit #1 | GPT 5.6 Sol openai Operational reference: GPT-5.6 Sol (max) ↗Active configuration | 63.0relative task estimate | $4.35per 1M blended | 55.9 tok/sexcludes first-token wait | 55% | 3 signal groups · 3 lineages EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| #12fit #4 | Claude Opus 4.8 anthropic Operational reference: Claude Opus 4.8 (max) ↗Active configuration | 61.2relative task estimate | $3.85per 1M blended | 54.8 tok/sexcludes first-token wait | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| #13fit #6 | Claude Opus 4.7 anthropic Operational reference: Claude Opus 4.7 (max) ↗Active configuration | 60.7relative task estimate | $3.85per 1M blended | 48.3 tok/sexcludes first-token wait | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| #14fit #10 | Kimi K2.6 moonshotai Operational reference: Kimi K2.6 ↗Active configuration | 58.5relative task estimate | $0.70per 1M blended | 44.4 tok/sexcludes first-token wait | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| —fit #5legacy · unranked | GPT 5.4 openai Legacy operational reference: GPT-5.4 (xhigh) ↗Deprecated · legacy data | 60.9relative task estimate | $2.17legacy reference · excluded from filters | 150.3 tok/slegacy reference · excluded from filters | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| —fit #7legacy · unranked | Claude Opus 4.6 anthropic Legacy operational reference: Claude Opus 4.6 (max) ↗Deprecated · legacy data | 60.4relative task estimate | $3.85legacy reference · excluded from filters | 39.0 tok/slegacy reference · excluded from filters | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| —fit #8legacy · unranked | Claude Sonnet 4.6 anthropic Legacy operational reference: Claude Sonnet 4.6 (max) ↗Deprecated · legacy data | 59.4relative task estimate | $2.31legacy reference · excluded from filters | 46.9 tok/slegacy reference · excluded from filters | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| —fit #11legacy · unranked | GPT 5.2 openai Legacy operational reference: GPT-5.2 (xhigh) ↗Deprecated · legacy data | 58.0relative task estimate | $1.87legacy reference · excluded from filters | 65.6 tok/slegacy reference · excluded from filters | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| —fit #13legacy · unranked | GPT 5 openai Legacy operational reference: GPT-5 (high) ↗Deprecated · legacy data | 57.3relative task estimate | $1.34legacy reference · excluded from filters | 99.0 tok/slegacy reference · excluded from filters | 55% | 3 signal groups · 3 lineages EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| —fit #14no operational data | Gemini 3 Pro google | 57.2relative task estimate | — | — | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| —fit #17no operational data | Deepseek V3.2 deepseek | 56.1relative task estimate | — | — | 55% | 3 signal groups · 3 lineages EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| —fit #19no operational data | Deepseek V3.1 deepseek | 54.9relative task estimate | — | — | 55% | 3 signal groups · 3 lineages EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| —fit #20legacy · unranked | Kimi K2.5 moonshotai Legacy operational reference: Kimi K2.5 ↗Deprecated · legacy data | 54.9relative task estimate | $0.49legacy reference · excluded from filters | 49.8 tok/slegacy reference · excluded from filters | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| —fit #22legacy · unranked | GPT 4.1 openai Legacy operational reference: GPT-4.1 ↗Deprecated · legacy data | 52.7relative task estimate | $1.55legacy reference · excluded from filters | 96.0 tok/slegacy reference · excluded from filters | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| —fit #23no operational data | DeepSeek-R1-0528 deepseek | 51.8relative task estimate | — | — | 55% | 3 signal groups · 3 lineages EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| —fit #24no operational data | DeepSeek-R1 deepseek | 51.5relative task estimate | — | — | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| —fit #25no operational data | DeepSeek-V3-0324 deepseek | 50.4relative task estimate | — | — | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| —fit #26no operational data | gemma-3-27b-it google | 49.6relative task estimate | — | — | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| —fit #27no operational data | gemma-3-12b-it google | 49.0relative task estimate | — | — | 55% | 3 signal groups · 3 lineages EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| —fit #28no operational data | qwq-32b alibaba | 47.4relative task estimate | — | — | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
| —fit #29no operational data | gemma-3-4b-it google | 46.4relative task estimate | — | — | 65% | 3 signal groups · 4 lineages EQ-Bench 3, EQ-Bench Creative Writing v3, LMArena Leaderboard Dataset, UGI Leaderboard |
The provisional fit order is useful for broad comparison, but small numerical differences are not evidence of precise gaps. Predicted-fit scores are meaningful only within this task page. †Operational values use the exact Artificial Analysis reference configuration linked on each row. Operational coverage: 14 active, 7 legacy, and 10 missing out of 31 model families. Only active references receive cost or throughput positions; active price coverage is 14/31 and active throughput coverage is 14/31 in the snapshot dated 2026-07-16. Read the complete methodology →
Publication threshold
A model appears in this beta only after meeting the category's shared evidence floor. For this release that means at least 55% signal coverage, 3 independent signal groups, and 3 evidence lineages. Lower-coverage models are withheld rather than assigned misleading positions.
Frequently asked
What is the best AI model for business writing?
GPT 5.6 Sol currently has the highest predicted-fit beta score. This is a relative estimate from several external benchmark signals, not a task-success percentage or a new direct model run.
Does output throughput mean fastest response?
No. Tokens per second measures generation after output begins and excludes time to first token and reasoning delay. Only active reference configurations receive a throughput position; deprecated references remain visible as legacy data. Throughput contributes no quality points.
How is predicted fit calculated?
SpringPrompt normalizes relevant third-party benchmark signals, combines them using reviewed task-specific weights, and keeps evidence coverage separate from performance. Only models meeting the published coverage and independent-evidence floors appear in the beta order.
How precise is the order?
It is a non-strict provisional order. Small score differences should be read as the same performance band, and scores should not be compared across different task pages.
How this beta is calculated
We normalize relevant public benchmark signals within their source, combine them using a reviewed category-specific weighting, and publish evidence coverage alongside the estimate. Missing signals remain missing. Product-family matching follows reviewed identity rules and does not create links to runnable models unless that identity is independently established.
Definition 2026-07-16-business-candidate-v6 · Release external-predictive-public-beta:fd72878b034342eaddb21f5bd557b0c000c016b7194764c76bc5dbd0b3384bae · sources, weights and scoring safeguards