Kimi K3 vs DeepSeek V4.1 Flash: pricing, allowances and benchmarks
Kimi K3 (Kimi) and DeepSeek V4.1 Flash (DeepSeek): 5 shared subscriptions and 6 shared leaderboard snapshots, with workloads, sample dates and benchmark configurations.
Plan & benchmark data snapshot · How real prices are computed
As of 2026-10-16, 5 of 5 shared subscriptions use matching normalized workloads: Kimi K3 has a lower recorded token cost on 0, DeepSeek V4.1 Flash on 5, with 0 tied. Their highest archived configurations score 43.59 and 39.46 on Artificial Analysis Intelligence Index v4.3; all 6 shared leaderboard snapshots, configurations and sources appear below.
- Shared subscriptions
- 5Both models available
- Comparable workloads
- 5Used for price ratios
- Shared leaderboards
- 6Highest archived configurations
- Official API prices
- 2/ 2Models with list prices
Plot both on the price–capability chart
Same plan, side by side
Each row is a subscription that includes both models at the same fee. Real cost = monthly fee ÷ the recorded monthly allowance. Each model's allowance assumes using the plan for that model; the columns cannot be added together. Click a price for its calculation and sources. Ratios use only matching normalized workloads.
| Plan / provider | Monthly fee | Kimi K3 | DeepSeek V4.1 Flash | Recorded token cost |
|---|---|---|---|---|
| Command Code GOATCommand Code | $10/mo | $0.2203 /MTok 0.045B tokens/month Standard workload 2026-09-30 · official source date · confidence Medium | $0.00161 /MTok 6.211B tokens/month Standard workload 2026-09-30 · official source date · confidence Medium | DeepSeek V4.1 Flash lower136.8× price ratio (higher/lower) |
| OpenCode GoOpenCode | $10/mo | $0.2941 /MTok 0.034B tokens/month Standard workload 2026-09-30 · official source date · confidence Medium | $0.00161 /MTok 6.211B tokens/month Standard workload 2026-09-30 · official source date · confidence Medium | DeepSeek V4.1 Flash lower182.7× price ratio (higher/lower) |
| Ollama ProOllama | $20/mo | $0.147 /MTok 0.136B tokens/month Standard workload 2026-09-06 · official source date · confidence Medium | $0.00322 /MTok 6.211B tokens/month Standard workload 2026-09-11 · official source date · confidence Medium | DeepSeek V4.1 Flash lower45.6× price ratio (higher/lower) |
| Ollama MaxOllama | $100/mo | $0.147 /MTok 0.68B tokens/month Standard workload 2026-09-06 · official source date · confidence Medium | $0.00322 /MTok 31.056B tokens/month Standard workload 2026-09-11 · official source date · confidence Medium | DeepSeek V4.1 Flash lower45.7× price ratio (higher/lower) |
| Droid MaxFactory | $200/mo | $0.05938 /MTok 3.368B tokens/month Anthropic workload 2026-09-29 · derived; uses the data date of Droid Max · claude-opus-5.5 · confidence Medium | $0.00594 /MTok 33.68B tokens/month Anthropic workload 2026-09-29 · derived; uses the data date of Droid Max · claude-opus-5.5 · confidence Medium | DeepSeek V4.1 Flash lower10× price ratio (higher/lower) |
Leaderboards head to head
Each model's highest archived configuration on each shared leaderboard snapshot, ranked among the models archived in this dataset. Harness, reasoning effort and service mode can differ; self-reported and estimated results are marked. These reference configurations are not measurements of the subscriptions above. Overlapping leaderboards do not form an overall ability rating.
| Leaderboard | Metric | Kimi K3 | DeepSeek V4.1 Flash | Higher archived score |
|---|---|---|---|---|
| Artificial Analysis Intelligence Index v4.3 | Intelligence Index | 43.59 #17 / 60 · Kimi K3 (max) Max 2026-10-01 · Record | 39.46 #25 / 60 · DeepSeek V4.1 Flash (max) Max 2026-10-01 · Record | Kimi K3+4.14 |
| Terminal-Bench 4.0 (AA) | Resolution Rate % | 12.63 #28 / 53 · Artificial Analysis - Kimi K3 (low) Artificial Analysis · Low 2026-10-01 · Record | 26.77 #17 / 53 · Artificial Analysis - DeepSeek V4.1 Flash (max) Artificial Analysis · Max 2026-10-01 · Record | DeepSeek V4.1 Flash+14.14 |
| Code Arena | WebDev Overall | Arena Score | 1,658 #10 / 54 · kimi-k3-max Max Reported interval 1,651.6–1,664.4 2026-10-01 · Record | 1,620 #16 / 54 · deepseek-v4.1-flash-max Max Reported interval 1,609.6–1,630.4 2026-10-01 · Record | Kimi K3+38 |
| Agent Arena · Overall (Ranking / Pareto) | Net Improvement % | 3.98 #14 / 39 · Kimi K3 (Max) Max Reported interval 3.45–4.51 2026-10-01 · Record | 4.01 #13 / 39 · Deepseek V4.1 Flash (Max) Max Reported interval 3.51–4.51 2026-10-01 · Record | DeepSeek V4.1 Flash+0.03 |
| OpenDesign Arena · Quality ranking | Average task score | 65.7 #16 / 16 · Kimi K3 OpenDesign 2026-10-01 · Record | 81.2 #8 / 16 · DeepSeek V4.1 Flash OpenDesign 2026-10-01 · Record | DeepSeek V4.1 Flash+15.5 |
| DeepSWE v1.1 | Pass@1 % | 68.51 #11 / 33 · mini-swe-agent - kimi-k3 (max) mini-swe-agent · Max Reported interval 63.98–73.05 2026-09-12 · Record | 74.2 self-reported #1 / 33 · mini-SWE - DeepSeek V4.1 Flash (max) mini-SWE · Max 2026-09-12 · Record | DeepSeek V4.1 Flash+5.69 |
Official API pricing
Each row uses its listed pricing conditions. Cache writes are shown separately; subscription costs above retain their recorded quota basis. An em dash means that rate is not recorded.
| Model / pricing conditions | Cached input | Input | Cache write | Output |
|---|---|---|---|---|
| Kimi K3Upstream referenceUpstream price reference | $0.3 | $3 | — | $15 |
| DeepSeek V4.1 FlashUpstream referenceUpstream price reference | $0.006 | $0.3 | — | $1.2 |
How to read this comparison
Price ratios compare recorded USD/MTok under the same known workload. Raw measured totals, different workloads and unmetered access remain visible but are excluded from the lower-cost counts. Recorded allowances assume using the quota; your effective cost depends on utilization, cache mix, output mix and plan limits. Qualitative confidence is not an error bar.
- Standard workload: 97% cache reads / 2.5% input / 0.5% output
- Anthropic workload: 97% cache reads / 2.5% cache writes / 0.5% output