Direct-API and provider pricing for current frontier & value LLMs — USD per 1M tokens.
| Model | Tier | Context | Max out | Best price ($/M in / out) | Notes |
|---|---|---|---|---|---|
| DeepSeek V4.1 Flashcore
DeepSeek |
cheap | 1M | 384K | $0.0825 / $0.33 via Cheaper Inference · 5 providers |
New architecture family (552B MoE, causal encoder–decoder: 8B active input / 16B active output). Native vision, 1M ctx, 384K max out; benchmarks beat V4 Pro. Replaces the retired V4 Flash (0731) and V4 Pro. |
| Nube-Choice
Nube |
cheap | 1M | 384K | $0.286 / $0.858 via Nube · 1 provider |
Nube's auto-routing model; currently priced in line with its DeepSeek-V4-Flash route (subject to change as routing shifts). |
| Qwen3.8-Maxcore
Alibaba Qwen |
frontier | 1M | 131K | $1.75 / $5.25 via Cheaper Inference · 2 providers |
2.4T MoE (95B active), multimodal. Flagship of Qwen3.8 series. |
| Qwen3.7-Pluscore
Alibaba Qwen |
mid | 1M | 131K | $0.32 / $1.28 via Direct API · 1 provider |
Cost-effective workhorse in the Qwen3.7 series. |
| Qwen3.7-Flash
Alibaba Qwen |
cheap | 1M | 131K | $0.03 / $0.13 via Direct API · 1 provider |
Qwen's budget text tier, 1M ctx. Cheapest model on the board. |
| Qwen3.8-Flash
Alibaba Qwen |
cheap | 1M | 131K | $0.15 / $0.47 via Direct API · 1 provider |
Production API of the open-weights Qwen3.8-Flash-Next (125B/6B-active MoE, 51B N-gram embed, Qwen4-arch preview). 1M ctx; multimodal (image in). |
| Qwen3.8-27B
Alibaba Cloud (Qwen) |
mid | 0.262M | 262K | $0.045 / $0.32 via InferX · 4 providers |
Open-weight (Apache-2.0) dense VLM, 262K ctx. Served via Cloudflare/OpenRouter/Vercel. |
| GPT-5.6 Lunasuperseded
OpenAI |
cheap | 1.05M | 128K | $0.08 / $0.48 via Cheaper Inference · 2 providers |
OpenAI's cheapest GPT-5.6 tier. Fast, low-cost general model. |
| GPT-5.6 Terra
OpenAI |
mid | 1.05M | 128K | $0.80 / $4.80 via Cheaper Inference · 2 providers |
Mid-tier GPT-5.6 (Terra). Balances cost and capability. |
| Claude Opus 5superseded
Anthropic |
frontier | 1M | 64K | $3.25 / $16.25 via Cheaper Inference · 2 providers |
Anthropic's frontier reasoning model. |
| Claude Sonnet 5superseded
Anthropic |
mid | 1M | 128K | $1.30 / $6.51 via Cheaper Inference · 2 providers |
Balanced mid-tier Claude for agents and tool use. The $2/$10 launch price, originally introductory through Aug 31 2026, is now standard. |
| GLM-5.3
Zhipu AI |
frontier | 1M | 131K | $0.63 / $1.98 via Nube · 3 providers |
Zhipu flagship coding/cyber model, 1M ctx. |
| GLM-5.3-Flash
Zhipu AI |
cheap | 1M | 131K | $0.06 / $0.20 via Cheaper Inference · 4 providers |
Natively-multimodal (first in GLM-5 series), 320B/18B active, 1M ctx. |
| MiniMax M3
MiniMax |
cheap | 1M | 524K | $0.30 / $1.20 via Direct API · 2 providers |
Multimodal foundation model; very cheap per token. 512K max out (128K recommended). |
| MiMo-V2.6-Pro
Xiaomi |
mid | 1.05M | 131K | $0.435 / $0.87 via Direct API · 1 provider |
Xiaomi's flagship omni-modal reasoning model (1T total / 42B active). AA Intelligence Index 46 — top open-weights score, above Kimi K3 and Qwen3.8-Max. V2.6 keeps V2.5 pricing. |
| MiMo-V2.6-Flash
Xiaomi |
cheap | 1.05M | 131K | $0.14 / $0.28 via Direct API · 1 provider |
Xiaomi's low-cost omni-modal reasoning model; replaces MiMo-V2.5 at the same rate with a 1M ctx. Strong for coding agents and long-context work. |
| Qwen3.7-Max
Alibaba Qwen |
frontier | 1M | 131K | $1.25 / $3.75 via Direct API · 1 provider |
Previous Qwen flagship; promotional pricing active. |
| GPT-5.6 Solsuperseded
OpenAI |
frontier | 1.05M | 128K | $2.00 / $10.00 via Cheaper Inference · 2 providers |
OpenAI's previous frontier general model. |
| Kimi K3
Moonshot AI |
frontier | 1M | 131K | $1.95 / $9.75 via Nube · 4 providers |
Newest Kimi flagship, 1M ctx. 128K max out default (raisable up to ~1M, shared ctx). |
| Claude Opus 5.5
Anthropic |
frontier | 1M | 128K | $4.00 / $20.00 via Direct API · 1 provider |
Anthropic's flagship agentic-coding / knowledge-work model. 1M ctx, 128K max out. Cheaper than the Opus 5 it replaces. Cache reads are 5% of input (a deeper discount than Anthropic's standard 10%). |
| Claude Sonnet 5.5
Anthropic |
mid | 1M | 128K | $2.00 / $10.00 via Direct API · 1 provider |
Mid-tier Claude successor to Sonnet 5 at the identical rate card, with a newer tokenizer. Same price as GPT-6.1 Sol but a 2x higher cache-read rate. |
| Claude Fable 5.1
Anthropic |
frontier | 1M | 128K | $10.00 / $50.00 via Direct API · 1 provider |
For demanding reasoning and long-horizon agentic work. Cache reads are 2.5% of input — the deepest cache discount Anthropic publishes. |
| GPT-6 Astra
OpenAI |
frontier | 1.05M | 128K | $10.00 / $50.00 via Direct API · 1 provider |
OpenAI's most capable model, for the hardest end-to-end work. Long-context (>272K input) bills at double these rates. |
| GPT-6.1 Sol
OpenAI |
frontier | 1.05M | 128K | $2.00 / $10.00 via Direct API · 1 provider |
"Near-Astra intelligence for a fifth of the price." Keeps GPT-6 Sol's $2/$10 rate card but halves cached input to $0.10 (5% of input) — half of Claude Sonnet 5.5's cache rate. Shipped a week after GPT-6 Sol, which it supersedes. GPT-6.1 Astra was shelved. |
| GPT-6 Luna
OpenAI |
cheap | 1.05M | 128K | $0.10 / $0.50 via Direct API · 1 provider |
OpenAI's most efficient model for focused, high-volume tasks. Half the input and output price of GPT-5.6 Luna with a 2x lower cache-read rate. |
| Model | Vendor | Est. monthly cost | Cost / 1M out | Cache / 1M in | Tier |
|---|
The cache-hit discount is applied per channel, using that channel's own published cache-read rate. The Cache / 1M in column shows the rate actually used; a dash means the selected channel publishes no cache-read price for that model, so every input token — cached or not — is billed at the full input rate and the slider will not move that row.
0% below the Alibaba Qwen direct rate of $0.03/$0.13.
Direct price — the cheapest input on the board.
The value flagship. GLM-5.3 ($1.40/$4.40) is a strong coding/cyber pick; Claude Opus 5 and GPT-5.6 Sol ($5/$25–$30) are the premium ceiling.
Z.ai list is $0.15/$0.50 (promo ended Sep 9). MiMo-V2.6-Flash matches the retired MiMo-V2.5 at $0.14/$0.28 with a 1M context.
Paired with GLM-5.3 at $0.63/$1.98 on Nube for the heavier coding and agent workloads.
DeepSeek V4.1 Flash, Qwen3.8, GPT-5.6 and GLM-5.3 all offer ~1M-token context.
Direct = the vendor's own API list price. Provider columns (Nube / Cheaper Inference, plus Fireworks / InferX whenever a vendor posts a rate for that model) = each provider's posted rate. Prices are USD per 1M tokens; DeepSeek shown at peak with cache-hit input pricing — off-peak is 50% lower. Cache-hit (cached-input) rates in the calculator are applied per selected channel, from that channel's own published cache-read price (vendor docs for Direct, each marketplace's catalog otherwise); where a channel publishes none, cached tokens are billed at the full input rate. Cache-write fees, batch pricing and per-vendor cache TTL/minimum-prefix rules are not modelled. Promotional prices are marked. Figures were hand-verified against official docs and provider pricing pages on the generation date. Latency/speed (TTFT, tokens/sec) = median figures from Artificial Analysis, refreshed live on each generation.