# Finance and accounting leaderboard

As of November 21, 2026, Gemini 4 Carbon leads our finance and accounting leaderboard at 89.2 (Estimated, $2/$10 per 1M tokens), ahead of Gemini 4 Neon at 87.6. All 35 rows currently carry Estimated labels; Kimi K3.1 Pro leads open-weight models at #10 (70.9). We rank 35 models across 8 domain protocols, and this vertical carries zero weight in the overall score.

## Shortlist by workload and budget

- **Highest point score:** [Gemini 4 Carbon](/models/gemini-4-carbon) (#1, 89.2, Estimated, $2/$10 per 1M tokens) — Carries an Estimated label because independent multi-evaluator coverage on finance & accounting is still thin.
- **Closest contender:** [Gemini 4 Neon](/models/gemini-4-neon) (#2, 87.6, Estimated, $3/$15 per 1M tokens) — Sits 1.6 points behind Gemini 4 Carbon; test both on your own prompts when provider lock-in or latency matters.
- **Top open-weight model:** [Kimi K3.1 Pro](/models/kimi-3-1-pro) (#10, 70.9, Estimated, Open weights · $2/$10 per 1M tokens) — Trails Gemini 4 Carbon by 18.3 points (#10 overall here), so pick it when weight control or on-prem residency outweighs peak score.

## Ranked models

| Rank | Model | Creator | Score | Conditional range | Evidence | API ($/1M in/out) | Context | Source families |
|---:|---|---|---:|---:|---|---:|---:|---:|
| #1 | [Gemini 4 Carbon](/models/gemini-4-carbon) | Google | 89.2 | 78.0-100.0 | Estimated | $2.00 / $10.00 | 1M | 1 |
| #2 | [Gemini 4 Neon](/models/gemini-4-neon) | Google | 87.6 | 75.6-99.7 | Estimated | $3.00 / $15.00 | 1M | 1 |
| #3 | [Gemini 4 Argon](/models/gemini-4-argon) | Google | 87.4 | 76.2-98.5 | Estimated | $2.00 / $10.00 | 1M | 1 |
| #4 | [Claude Fable 5.2](/models/claude-fable-5-2) | Anthropic | 81.8 | 70.4-93.2 | Estimated | $8.00 / $40.00 | 1M | 2 |
| #5 | [Claude Opus 5.5](/models/claude-opus-5-5) | Anthropic | 77.4 | 66.6-88.2 | Estimated | $4.00 / $20.00 | 1M | 2 |
| #6 | [Gemini 4 Hydrogen](/models/gemini-4-hydrogen) | Google | 79.1 | 68.0-90.3 | Estimated | $1.50 / $7.50 | 1M | 1 |
| #7 | [GPT-6.2 Astra](/models/gpt-6-2-astra) | OpenAI | 78.4 | 66.3-90.4 | Estimated | $8.00 / $40.00 | 1.05M | 1 |
| #8 | [Muse Wonder](/models/muse-wonder) | Meta | 74.9 | 63.4-86.4 | Estimated | $4.00 / $20.00 | 1M | 1 |
| #9 | [Claude Fable 5.1](/models/claude-fable-5-1) | Anthropic | 74.6 | 59.3-90.0 | Estimated | $10.00 / $50.00 | 1M | 1 |
| #10 | [Kimi K3.1 Pro](/models/kimi-3-1-pro) | Moonshot AI | 70.9 | 55.2-86.6 | Estimated | $2.00 / $10.00 | 1.05M | 2 |
| #11 | [GPT-6.2 Sol](/models/gpt-6-2-sol) | OpenAI | 72.1 | 59.4-84.8 | Estimated | $2.00 / $10.00 | 1.05M | 1 |
| #12 | [Claude Opus 5](/models/claude-opus-5) | Anthropic | 71.3 | 56.0-86.5 | Estimated | $5.00 / $25.00 | - | 1 |
| #13 | [Claude Sonnet 5.5](/models/claude-sonnet-5-5) | Anthropic | 68.8 | 58.0-79.7 | Estimated | $2.00 / $10.00 | 1M | 2 |
| #14 | [Hy4 preview](/models/hy4-preview) | Tencent | 68.2 | 51.9-84.6 | Estimated | Free | 1M | 1 |
| #15 | [GLM-5.3](/models/glm-5-3) | Z.AI | 68.1 | 57.0-79.2 | Estimated | Free | 1M | 1 |
| #16 | [Claude Haiku 5.5](/models/claude-haiku-5-5) | Anthropic | 67.2 | 56.1-78.3 | Estimated | $0.10 / $0.50 | 1M | 1 |
| #17 | [Inkling 2 Small](/models/inkling-2-small) | Thinking Machines Lab | 66.0 | 43.6-88.5 | Estimated | $0.30 / $1.00 | 1M | 1 |
| #18 | [Qwen3.8 Max](/models/qwen3-8-max) | Alibaba | 66.0 | 55.0-77.1 | Estimated | - | 1M | 1 |
| #19 | [Kimi K3](/models/kimi-k3) | Moonshot AI | 64.0 | 52.9-75.2 | Estimated | $3.00 / $15.00 | 1.05M | 2 |
| #20 | [Claude Fable 5](/models/claude-fable) | Anthropic | 65.9 | 50.2-81.7 | Estimated | $10.00 / $50.00 | 1M+ | 1 |
| #21 | [DeepSeek V4.1 Flash 1017](/models/deepseek-v4-1-flash) | DeepSeek | 65.9 | 54.8-77.0 | Estimated | $0.30 / $1.20 | 1M | 1 |
| #22 | [Claude Opus 4.8](/models/claude-opus-4-8) | Anthropic | 65.7 | 50.3-81.0 | Estimated | $5.00 / $25.00 | 1M | 1 |
| #23 | [Step 5 Preview](/models/step-5-preview) | StepFun | 61.5 | 43.8-79.2 | Estimated | $1.00 / $2.70 | 1M | 1 |
| #24 | [GLM-5.3-Flash](/models/glm-5-3-flash) | Z.AI | 58.2 | 47.1-69.3 | Estimated | Free | 1M | 2 |
| #25 | [GPT-5.5](/models/gpt-5-5) | OpenAI | 59.2 | 43.6-74.9 | Estimated | $5.00 / $30.00 | 1M | 1 |
| #26 | [GPT-5.4](/models/gpt-5-4) | OpenAI | 56.2 | 40.8-71.7 | Estimated | $2.50 / $15.00 | 1.05M | 1 |
| #27 | [GPT-6 Luna](/models/gpt-6-luna) | OpenAI | 54.1 | 43.0-65.3 | Estimated | $0.10 / $0.50 | 1.05M | 1 |
| #28 | [Claude Opus 4.7](/models/claude-opus-4-7) | Anthropic | 49.6 | 32.0-67.2 | Estimated | $5.00 / $25.00 | 1M | 1 |
| #29 | [Mixtral 8x22B Instruct v0.1](/models/mixtral-8x22b-instruct-v0-1) | Mistral | 43.5 | 10.1-76.9 | Estimated | Free | 64K | 4 |
| #30 | [MiniMax M3](/models/minimax-m3) | MiniMax | 44.6 | 28.9-60.3 | Estimated | $0.30 / $1.20 | 1M | 1 |
| #31 | [Gemini 1.5 Pro](/models/gemini-1-5-pro) | Google | 40.9 | 19.5-62.3 | Estimated | $1.25 / $5.00 | 2M | 4 |
| #32 | [Claude 3.5 Sonnet](/models/claude-3-5-sonnet) | Anthropic | 40.3 | 20.6-60.1 | Estimated | $3.00 / $15.00 | 200K | 4 |
| #33 | [GPT-4o](/models/gpt-4o) | OpenAI | 37.7 | 17.0-58.5 | Estimated | $2.50 / $10.00 | 128K | 4 |
| #34 | [Claude 3 Opus](/models/claude-3-opus) | Anthropic | 36.5 | 13.8-59.1 | Estimated | $15.00 / $75.00 | 200K | 4 |
| #35 | [GPT-4o mini](/models/gpt-4o-mini) | OpenAI | 33.4 | 14.2-52.6 | Estimated | $0.15 / $0.60 | 128K | 4 |

## Scored domain benchmarks (8)

- [Ramp Accounting Bench](/benchmarks/ramp-accounting-bench) (33% ref. weight): A Ramp Labs public leaderboard for long-horizon accounting agent work. BenchLM shows it as display-only reference data and excludes it from model rankings.
- [OfficeQA Pro](/benchmarks/officeqapro) (22% ref. weight): A benchmark for grounded reasoning over office-style documents, spreadsheets, charts, and business artifacts.
- [Diligence Bench](/benchmarks/diligencebench) (10% ref. weight): A financial due-diligence agent benchmark from Thoughtful, scoring how well a model researches and analyzes a company the way an investment analyst would.
- [Frontier Finance](/benchmarks/frontierfinance) (10% ref. weight): An investment-workflow benchmark covering complete tasks from screening to profitability analysis.
- HELM Finance: BANKING77 · Stanford CRFM HELM Finance (6% ref. weight): Fine-grained retail banking intent classification across 77 customer-service categories.
- HELM Finance: FinanceBench · Stanford CRFM HELM Finance (6% ref. weight): Open-book financial question answering over public 10-K, 10-Q, and earnings filings.
- HELM Finance: FinQA · Stanford CRFM HELM Finance (6% ref. weight): Multi-step numerical reasoning over corporate financial tables and footnotes.
- Stanford HELM Finance (win rate) · Stanford CRFM HELM Finance (6% ref. weight): Aggregate head-to-head win rate across the Apache-2.0 Stanford HELM Finance suite.

## Ledger reconciliation, SEC filing arithmetic, and institutional banking tasks

Seven protocols set the order on this board, led by Ramp Accounting Bench (33%), OfficeQA Pro (22%), Diligence Bench (10%) plus four Apache-2.0 Stanford CRFM HELM Finance evaluations (FinQA, FinanceBench, BANKING77, and HELM Finance win rate). Ramp Accounting Bench grades multi-step ledger categorization, journal entries, and reconciliation rules, while OfficeQA Pro and FinQA test whether models compute exact ratios and period-over-period variances from corporate filings without rounding drift.

Because BenchAlign v6.0 calibrates each benchmark's difficulty against our 92-model reference cohort, models with only easy intent-classification rows cannot leapfrog models tested on full accounting workflows.

## Why every row is currently Estimated—and what finance benchmarks never certify

All 35 ranked models on this board carry Estimated labels (0 Supported) because independent multi-evaluator finance coverage across the full frontier catalog is still thinner than on coding or knowledge. Expect rank swaps among closely spaced models as additional open financial evaluations land.

More importantly, benchmark accuracy is not GAAP or IFRS audit signoff, tax compliance, or investment advice. Models that score near 89 here can still transpose a footnote unit between thousands and millions or misread an amended restatement; wire deterministic Python or spreadsheet execution into any production calculation path.

[Browse all finance benchmark tables](/benchmarks/collections/finance) · [Gemini 4 Carbon vs Gemini 4 Neon](/compare/gemini-4-carbon-vs-gemini-4-neon)

## Selecting models for diligence memos, spreadsheet agents, and on-prem banking

Gemini 4 Carbon (89.2, $2/$10 per 1M tokens) and Gemini 4 Neon (87.6, $3/$15 per 1M tokens) sit separated by 1.6 points, ahead of Gemini 4 Argon (87.4). Because Gemini 4 Carbon lists at $2/$10 per 1M tokens, teams running SEC filing extraction get the highest current finance estimate without paying top-tier $8/$40 pricing, while mid-tier models handle routine classification cheaply.

Regulated institutions keeping client ledgers inside private infrastructure can shortlist Kimi K3.1 Pro (#10, 70.9), paired with tool-executed arithmetic so the model never performs unverified floating-point math in prose.

[Math leaderboard](/math) · [Documents & extraction leaderboard](/leaderboards/documents)

## Questions

### Which LLM is best for finance and accounting right now?

Gemini 4 Carbon holds #1 on our finance and accounting leaderboard at 89.2 (Estimated) as of November 21, 2026, ahead of Gemini 4 Neon at 87.6 and Gemini 4 Argon at 87.4. Every ranked row currently carries an Estimated label, so validate top contenders on your own workload.

### What is the best open-weight LLM for finance and accounting?

Kimi K3.1 Pro from Moonshot AI leads open-weight models on this finance and accounting board at #10 with a calibrated score of 70.9 (Estimated). That trails Gemini 4 Carbon by 18.3 points, so choose Kimi K3.1 Pro when self-hosting, VPC data residency, or weight customization matters more than peak frontier accuracy.

### Which API model offers the best value for finance and accounting?

Gemini 4 Carbon leads both point accuracy (89.2) and price efficiency here at $2/$10 per 1M tokens, undercutting $8/$40 flagship tiers without sacrificing finance and accounting quality. If you need an even lower input rate for high-volume batch runs, compare mid-tier alternatives in the table below against your own retry rate.

### Which benchmarks make up the finance and accounting leaderboard score?

BenchAlign v6.0 calibrates 8 open and primary domain protocols for this board, led by Ramp Accounting Bench (33%), OfficeQA Pro (22%), Diligence Bench (10%), Frontier Finance (10%). Each percentage reflects share of the vertical's reference weight inside our item-response model rather than a raw score average, and no single independent evaluator can exceed 40% when three or more evaluators contribute.

### Why do all finance leaderboard rows carry an Estimated label right now?

All 35 finance rows are labeled Estimated because independent multi-evaluator coverage across Ramp Accounting Bench, OfficeQA Pro, Frontier Finance, and Stanford HELM Finance does not yet clear our Supported source-family threshold for individual models. Rankings remain ordered by calibrated point estimates, but close gaps warrant testing on your own filings.

### Can a top-ranked finance LLM sign off on accounting or tax filings?

No. Ramp Accounting Bench, FinQA, and OfficeQA Pro measure structured bookkeeping workflows and filing arithmetic under test conditions, not GAAP or IFRS audit compliance, fiduciary suitability, or tax law judgment. Route model arithmetic through deterministic code execution and require review by a qualified accountant or analyst.

## Related benchmarks and guides

- [Finance Benchmark Collection](/benchmarks/collections/finance): Compare per-benchmark protocols across accounting, filings, tax, and coaching.
- [Math Leaderboard](/math): Symbolic and numerical reasoning rankings across 13 math benchmarks.
- [Documents & Extraction Leaderboard](/leaderboards/documents): PDF table parsing and structured schema extraction rankings.
- [LLM API Pricing](/llm-pricing): Compare token costs for batch filing analysis and diligence pipelines.

## Other specialist vertical leaderboards

- [Cybersecurity leaderboard](/leaderboards/cybersecurity) — #1 Gemini 4 Carbon (90.0)
- [Documents and extraction leaderboard](/leaderboards/documents) — #1 GPT-6.2 Astra (86.5)
- [Factuality and anti-hallucination leaderboard](/leaderboards/factuality) — #1 Gemini 4 Carbon (78.8)
- [Frontend and web development leaderboard](/leaderboards/frontend) — #1 Claude Fable 5.2 (89.5)
- [Healthcare and medical leaderboard](/leaderboards/healthcare) — #1 Claude Fable 5.2 (87.5)
- [Legal leaderboard](/leaderboards/legal) — #1 Claude Fable 5.2 (86.1)
- [Science and research leaderboard](/leaderboards/science) — #1 Claude Fable 5.2 (94.1)
- [Search and deep research leaderboard](/leaderboards/search) — #1 Claude Fable 5.2 (86.6)

Last updated: November 21, 2026

Canonical page: https://benchlm.ai/leaderboards/finance
