The OpenRouter list price for flagship models has fallen sharply since 2024. Claude's top Opus tier went from $75 to $20 per 1M output tokens, and OpenAI's line moved from GPT-4o at $5 / $15 to GPT-6 Sol at $2 / $10. Prices still swing often: in the last two weekly change logs, many models moved in both directions, and some reversed within a week.
This article uses OpenRouter list prices from weekly snapshots held by the Internet Archive, covering 88 weekly snapshots since 2024-03-26. All prices are USD per 1M tokens (input / output) unless stated.
List price is not task cost
The list price is what a developer sees on OpenRouter. A task's cost also depends on how many tokens the model reads, how many it writes, and how many reasoning tokens it spends first. Reasoning tokens are billed as output tokens, so a cheap output price can still produce an expensive task. The timelines below show list price. The last section uses my own task measurements (see the methodology).
Claude Opus: from $75 to $20 output
Claude 3 Opus was first seen on 2024-03-26 at $15 input and $75 output. Opus 4 and Opus 4.1 kept the same prices. The step down came with Opus 4.5, then Opus 5.5 cut it again.
| Model | First seen | Input | Output |
|---|---|---|---|
| Claude 3 Opus | 2024-03-26 | $15 | $75 |
| Claude Opus 4 | 2025-06-05 | $15 | $75 |
| Claude Opus 4.1 | 2025-08-12 | $15 | $75 |
| Claude Opus 4.5 | 2025-11-30 | $5 | $25 |
| Claude Opus 5 | 2026-07-28 | $5 | $25 |
| Claude Opus 5.5 | 2026-09-29 | $4 | $20 |
Sonnet and Haiku moved less. Claude 3 Sonnet started at $3 / $15, and Claude Sonnet 5.5 was first seen on 2026-09-29 at $2 / $10. Claude Haiku 5.5, first seen 2026-10-11, lists at $0.10 / $0.50, against $0.25 / $1.25 for Claude 3 Haiku in 2024.
Not everything fell. Claude 3.5 Sonnet's latest list price is $6 / $30 (2026-03-30), double its original $3 / $15.
OpenAI: GPT-4o to GPT-6
GPT-4o was first seen on 2024-06-19 at $5 / $15 and was listed at $2.5 / $10 on 2026-10-11. GPT-4o-mini was first seen at $0.15 / $0.6.
The GPT-6 generation does not mean one price:
- GPT-6 Luna (first seen 2026-09-29): $0.1 / $0.5
- GPT-6 Sol (2026-09-29): $2 / $10
- GPT-6.1 Sol (2026-10-05): $2 / $10
- GPT-6 Astra (2026-09-07): $10 / $50
Astra's output price is five times Sol's. The GPT-5 line in between rose before it fell: GPT-5 (2025-08-12) was $1.25 / $10, GPT-5.5 (2026-04-27) was $5 / $30, and GPT-5.6 Sol started at $5 / $30 and was listed at $2 / $10 on 2026-10-11. Likewise o3 started at $10 / $40 and is now listed at $2 / $8.
DeepSeek: cheap, then volatile
DeepSeek-V2 Chat was first seen on 2024-06-19 at $0.14 / $0.28. Its latest list price is $0.32 / $0.89, so it got more expensive. DeepSeek R1 went from $0.55 / $2.19 to $0.7 / $2.5, and V3.2 from $0.27 / $0.4 to $0.259 / $0.8.
The V4 family moved most:
| Model | First seen | Starting price | Latest (2026-10-11) |
|---|---|---|---|
| DeepSeek V4 Flash | 2026-04-27 | $0.14 / $0.28 | $0.03 / $1.28 |
| DeepSeek V4 Pro | 2026-04-27 | $0.435 / $0.87 | $0.2088 / $0.4176 |
| DeepSeek V4 Pro 0813 | 2026-08-14 | $0.435 / $0.87 | $0.66 / $1.98 |
| DeepSeek V4.1 Flash | 2026-09-14 | $0.15 / $0.6 | $0.3 / $1.2 |
Recent increases and reversals
The long-run story is downward, but recent weeks are noisy. Examples from my change log:
- DeepSeek V4.1 Flash: on 2026-10-05 output rose from $1.20 to $2.40 (+100%) and input fell to $0.003. On 2026-10-11 it reversed: input back to $0.30, output back to $1.20 (-50%).
- GLM 5.2: output rose from $2.04 to $8.00 (+292%) on 2026-10-05, then fell to $7.00 (-12%) on 2026-10-11.
- GLM Flash Latest: output rose from $0.14 to $0.50 (+257%) on 2026-10-05.
- DeepSeek V4 Flash 0423: output rose from $0.28 to $1.28 (+357%) on 2026-10-05.
- Kimi K3: output fell from $15.00 to $14.00 on 2026-10-05, then rose back to $15.00 (+7%) on 2026-10-11.
- Claude Haiku Latest: on 2026-10-11 output fell from $5.00 to $0.50 (-90%).
One or two weeks is not a trend. What I can say is that providers adjust list prices often, so last month's price may not be today's.
The catalog got bigger
The OpenRouter list grew from 121 models on 2024-03-26 to 182 on 2024-10-01, 324 on 2025-05-03, 422 on 2026-08-24 and 458 on 2026-10-11. More choice makes cheap options easier to find, but comparisons harder, because one brand now ships several tiers at very different prices.
Cost per task tells a different story
Anthropic's tokenizer ratio is 1.39 against OpenAI's 1.00 for the same English document, so the same text becomes more tokens on Claude models. Reasoning adds more spread. In my chat answer test, the median of all 326 models is 0.24¢ per task.
| # | Model | Cost per task | Text score | Tokens in / out | Basis | |
|---|---|---|---|---|---|---|
| 1 |  GLM 5.3 FlashZ.ai | 0.04¢$0.36 / 1,000 | 1474 | 48 / 1,782 | measured best value | |
| 2 |  GPT-6 LunaOpenAI | 0.06¢$0.59 / 1,000 | 1443 | 45 / 1,179 | measured | |
| 3 |  Claude Haiku 5.5Anthropic | 0.07¢$0.66 / 1,000 | - | 61 / 1,299 | measured | |
| 4 |  Qwen3.8 FlashAlibaba Qwen | 0.08¢$0.76 / 1,000 | - | 88 / 1,598 | measured | |
| 5 |  Mistral Small 4Mistral | 0.08¢$0.83 / 1,000 | - | 40 / 1,377 | estimated | |
| 6 |  MiMo-V2.6-ProXiaomi | 0.09¢$0.90 / 1,000 | 1479 | 206 / 933 | estimated best value | |
| 7 |  DeepSeek V4.1 FlashDeepSeek | 0.09¢$0.91 / 1,000 | 1475 | 65 / 6,320 | measured | |
| 8 |  Llama 4 MaverickMeta | 0.11¢$1.06 / 1,000 | 1327 | 87 / 1,592 | estimated |
GPT-6 Luna costs 0.06¢ per task and Claude Fable 5.1 costs 6.3¢. Claude Opus 5.5 costs 2.5¢ and Claude Sonnet 5.5 costs 1.3¢. DeepSeek V4.1 Flash used 5,461 reasoning tokens and cost 0.76¢, while Claude Sonnet 5.5 used 0 reasoning tokens and cost 1.3¢. Cheap per-token pricing does not guarantee a cheap task.
To check your own workload, use the calculator or see the chatbot answer page. For tier comparisons, start with Claude Opus 5.5 and GPT-6 Luna.
What to take from this
Flagship declines are real: Opus output fell from $75 to $20, and GPT-4o's list price fell to $2.5 / $10. But plan against the current price, expect it to change, and measure tokens and reasoning on your own workload before committing.
FAQ
Did LLM prices really fall between 2024 and 2026?
For the flagships I tracked, yes. Claude Opus output fell from $75 to $20 per 1M tokens, and GPT-4o went from $5 / $15 to $2.5 / $10. Some models rose, including DeepSeek-V2 Chat, now listed at $0.32 / $0.89.
Why does the same model change price so often?
Providers adjust list prices frequently, and OpenRouter reflects it week to week. DeepSeek V4.1 Flash went from $1.20 to $2.40 output on 2026-10-05, then back to $1.20 on 2026-10-11. Check the current price before budgeting.
Is a cheap list price enough to pick a model?
No. Reasoning tokens are billed as output, and tokenizers differ by provider. DeepSeek V4.1 Flash used 5,461 reasoning tokens and cost 0.76¢ per chat answer, while Claude Sonnet 5.5 used none and cost 1.3¢.