Kimi K3 vs Claude Opus 5
Kimi K3 is marketed for long-horizon coding and knowledge work, the same ground as Opus. Both list prices from the makers’ own pages, and what three everyday jobs cost on each.
US dollars, standard rate, shortest context band. Price only: this page does not rank quality.
- Kimi K3
- $15.00
- Claude Opus 5
- $25.00
- Kimi K3, per M tokens
- $3 / $15
- Claude Opus 5, per M tokens
- $5 / $25
3,000 tokens in and 400 out each, at list price.
Per million tokens, Kimi K3 costs $3 in and $15 out; Claude Opus 5 costs $5 in and $25 out. 1,000 support replies cost $25.00 on Claude Opus 5 and $15.00 on Kimi K3, so Claude Opus 5 costs 67% more. Kimi K3 reads up to 1.05M tokens at once, Claude Opus 5 up to 1M.
Side by side
Each maker’s published rate for its standard tier. Cached input is what a repeated prompt prefix costs once the provider has stored it.
| Fact | Kimi K3 | Claude Opus 5 |
|---|---|---|
| Maker | Moonshot AI | Anthropic |
| Tier | Flagships | Flagships |
| Input, per M tokens | $3.00 | $5.00 |
| Output, per M tokens | $15.00 | $25.00 |
| Cached input | $0.30 | $0.50 |
| Context window | 1.05M | 1M |
| Released | Jul 2026 | 24 Jul 2026 |
| Weights | Open | Closed |
What a job costs on each
1,000 runs of each job at the two list prices. The gap moves from job to job because the models price input, output and cached context differently.
| Job | Kimi K3 | Claude Opus 5 | Cheaper |
|---|---|---|---|
| Support reply3,000 in, 400 out. | $15.00 | $25.00 | Kimi K3, 40% less |
| Document summary25,000 in, 1,000 out. | $90.00 | $150 | Kimi K3, 40% less |
| Agent task80,000 in (70,000 cached), 3,000 out. | $96.00 | $160 | Kimi K3, 40% less |
The token counts are ours, chosen to look like real work; the prices are the makers’. A model with no published cache price bills the agent task’s repeated context at its full input rate.
The small print
What each base rate leaves out, from the makers’ own pages: surcharges, discounts, promotions and cache pricing.
- Kimi K3
- Cache write $3 (5 min TTL) or $6 (1 h TTL). Open weights on Hugging Face (moonshotai/Kimi-K3). Changelog gives the month only. Moonshot AI’s pricing page
- Claude Opus 5
- Cache write $6.25 (5 min) or $10 (1 h). Fast mode $10 / $50. Batch API 50% off. US-only inference (inference_geo) bills 1.1x. Full context window billed at standard rates. Anthropic’s pricing page
Other comparisons
Every head to head on the list that involves Kimi K3 or Claude Opus 5.
Run both for a quarter of list price
AIGROW API credit covers Kimi K3 and Claude Opus 5 and is metered at the list prices above: $25 buys $100 of usage.
Prices checked 22 September 2026