Blog ·
DeepSeek V4.1 Flash 1017: every public price, compared
During the beta it costs nothing here, with no rate limit. After the beta: $0.005 per million input tokens, $0.001 per million cached input tokens and $0.14 per million output tokens. On 8 November 2026 that was below every one of the 51 paid offers we found, on each of the three prices.
The short version
- Lowest input price elsewhere: $0.006 (Relace). Ours: $0.005.
- Lowest cached input price elsewhere: $0.0015 (Particle AI). Ours: $0.001.
- Lowest output price elsewhere: $0.15 (Sail Research). Ours: $0.14.
- A coding-agent day (below) costs $0.14 here after the beta, against $0.304 at the cheapest other offer (Particle AI).
- DeepSeek's own API is listed at $0.15 input, $0.003 cached and $0.60 output, so the same day costs $1.07 there.
What a coding-agent day costs
Prices per token don't say much on their own, because coding agents (Claude Code, Codex, OpenCode) resend the whole conversation on every turn and most of that input is a cache hit. So the last column prices one heavy day: 40 million input tokens, 90% of them cached, and 600,000 output tokens.
| Offer | Input / 1M | Cached / 1M | Output / 1M | Agent day |
|---|---|---|---|---|
| cheapllm.shop (after the beta) | $0.005 | $0.001 | $0.14 | $0.14 |
| Particle AI (us/flex) | $0.025 | $0.0015 | $0.25 | $0.304 |
| Morph (fp8) | $0.022 | $0.004 | $0.40 | $0.472 |
| Relace (fp4) | $0.006 | $0.005 | $0.50 | $0.504 |
| DeepInfra (fp8/flex) | $0.078 | $0.0024 | $0.235 | $0.541 |
| Particle AI (us) | $0.05 | $0.003 | $0.50 | $0.608 |
| OpenInference (fp4) | $0.018 | $0.004 | $0.70 | $0.634 |
| DeepInfra (fp8) | $0.098 | $0.0029 | $0.294 | $0.673 |
| Baidu (fp8) | $0.098 | $0.0021 | $0.394 | $0.703 |
| Sail Research (fp4/us) | $0.101 | $0.0067 | $0.15 | $0.735 |
| Alibaba | $0.075 | $0.0075 | $0.30 | $0.75 |
| StreamLake (fp8) | $0.111 | $0.0022 | $0.444 | $0.79 |
| Wafer (mxfp4) | $0.066 | $0.0065 | $0.50 | $0.798 |
| Wafer (us) | $0.038 | $0.01 | $0.50 | $0.812 |
| Cloudflare | $0.10 | $0.003 | $0.60 | $0.868 |
| Decart (fp8) | $0.123 | $0.0025 | $0.492 | $0.877 |
| InferenceNet (us) | $0.08 | $0.008 | $0.55 | $0.938 |
| InferenceNet (fp8) | $0.08 | $0.008 | $0.55 | $0.938 |
| StepFun (fp8) | $0.135 | $0.0027 | $0.54 | $0.961 |
| DeepInfra (fp8/priority) | $0.147 | $0.0044 | $0.441 | $1.01 |
| DeepSeek | $0.15 | $0.003 | $0.60 | $1.07 |
| Sail Research (fp4) | $0.151 | $0.01 | $0.225 | $1.10 |
| Novita (fp8) | $0.165 | $0.0033 | $0.66 | $1.17 |
| PrimeIntellect | $0.087 | $0.0048 | $1.20 | $1.24 |
| GMICloud (fp8) | $0.18 | $0.0036 | $0.72 | $1.28 |
| ModelRun (fp4) | $0.17 | $0.006 | $0.67 | $1.30 |
| Ionstream | $0.165 | $0.011 | $0.495 | $1.33 |
| DekaLLM | $0.18 | $0.006 | $0.68 | $1.34 |
| DigitalOcean | $0.18 | $0.006 | $0.72 | $1.37 |
| AtlasCloud (fp8) | $0.195 | $0.0039 | $0.78 | $1.39 |
| Cohere (fp8) | $0.195 | $0.006 | $0.66 | $1.39 |
| Fireworks (spot) | $0.21 | $0.0042 | $0.84 | $1.50 |
| Makora (fp8) | $0.15 | $0.006 | $1.20 | $1.54 |
| Reka | $0.16 | $0.008 | $1.10 | $1.59 |
| PrimaLabs | $0.24 | $0.0048 | $0.96 | $1.71 |
| Phala | $0.20 | $0.006 | $1.20 | $1.74 |
| SiliconFlow (fp8) | $0.30 | $0.006 | $0.55 | $1.75 |
| Modal | $0.25 | $0.005 | $1.00 | $1.78 |
| Relace (fast) | $0.30 | $0.008 | $0.90 | $2.03 |
| Crusoe (fp8) | $0.29 | $0.007 | $1.20 | $2.13 |
| CoreWeave (fp8) | $0.30 | $0.006 | $1.20 | $2.14 |
| Near AI | $0.30 | $0.006 | $1.20 | $2.14 |
| Nebius (fp8) | $0.30 | $0.006 | $1.20 | $2.14 |
| Together | $0.30 | $0.006 | $1.20 | $2.14 |
| Fireworks | $0.30 | $0.006 | $1.20 | $2.14 |
| Parasail (fp8) | $0.30 | $0.006 | $1.20 | $2.14 |
| Mistral (fp8) | $0.30 | $0.007 | $1.20 | $2.17 |
| BaseTen | $0.30 | $0.007 | $1.20 | $2.17 |
| Venice (fp8) | $0.30 | $0.0075 | $1.20 | $2.19 |
| Fireworks (us) | $0.45 | $0.009 | $1.80 | $3.20 |
| BaseTen (fast) | $0.60 | $0.014 | $2.40 | $4.34 |
| Fireworks (ultrafast) | $2.40 | $0.048 | $9.60 | $17.09 |
How we collected this
The 51 offers are every paid endpoint for deepseek/deepseek-v4.1-flash-1017 in OpenRouter's public model listing on 8 November 2026, DeepSeek's own API included. Names in brackets are the variant each host lists (precision such as fp8 or fp4, or tiers such as "flex" and "priority"). Some prices are promotions; we used the listed price as it stood that day. Where a host lists no cached-input price, the agent day counts cached tokens at its input price.
Prices move. When they do, we take a new snapshot and rebuild this table.